图像质量评估方法及图像质量评估***Image quality assessment method and image quality evaluation system
本申请要求2017年9月8日提交的申请号为No.201710804804.0的中国申请的优先权,通过引用将其全部内容并入本文。The present application claims the priority of the Chinese application filed on Sep. 8, 2017, the entire disclosure of which is hereby incorporated by reference.
技术领域Technical field
本发明涉及图像处理技术领域,特别涉及一种图像质量评估方法及图像质量评估***。The present invention relates to the field of image processing technologies, and in particular, to an image quality evaluation method and an image quality evaluation system.
发明背景Background of the invention
随着光学字符识别(Optical Character Recognition,OCR)技术的广泛应用,利用OCR技术采集得到的文本图像的质量日益受到更多关注,针对于文本图像的质量评估方法也引起学术界和工业界的更广泛的兴趣。With the wide application of optical character recognition (OCR) technology, the quality of text images acquired by OCR technology has received more and more attention. The quality evaluation method for text images has also caused more academic and industrial circles. Wide interest.
现有图像质量评估方法主要可分为两大类,一类是有参考图像的质量评估,另一类是无参考图像的质量评估。其中,有参考图像的质量评估是将失真图像与原始图像进行特征(比如梯度、对比度等)对比,得到失真图像的质量评估;无参考图像的质量评估是直接提取失真图像的某些特征(比如边缘强度、模糊程度等),并根据所提取的特征得到失真图像的质量评估。现有图像质量评估方法存在算法复杂、计算量大的缺陷,并且大多数现有图像质量评估方法需要预先对待评估图像进行处理,处理后的待评估图像才能进行评估操作,评估过程较为复杂。此外,大多数现有图像质量评估方法针对的是自然场景图像,并不适合文本图像的质量评估。The existing image quality assessment methods can be mainly divided into two categories, one is the quality evaluation of the reference image, and the other is the quality assessment without the reference image. Among them, the quality evaluation of the reference image is to compare the distortion image with the original image (such as gradient, contrast, etc.) to obtain the quality evaluation of the distorted image; the quality evaluation of the non-reference image is to directly extract some features of the distorted image (such as Edge strength, degree of blur, etc.), and quality assessment of the distorted image is obtained based on the extracted features. Existing image quality assessment methods have the disadvantages of complex algorithms and large computational complexity, and most existing image quality assessment methods require pre-evaluation of the images to be processed, and the processed images to be evaluated can be evaluated. The evaluation process is more complicated. In addition, most existing image quality assessment methods are aimed at natural scene images and are not suitable for quality assessment of text images.
发明内容Summary of the invention
有鉴于此,本发明实施例提供一种图像质量评估方法及图像质量评估***,以解决现有图像质量评估方法评估精准性差、评估效率低的问题,尤其是针对于文本图像的评估操作。In view of this, the embodiment of the present invention provides an image quality evaluation method and an image quality evaluation system to solve the problem that the existing image quality evaluation method has poor accuracy and low evaluation efficiency, especially for the evaluation operation of the text image.
第一方面,本发明实施例提供一种图像质量评估方法,该图像质量评估方法包括利用图像样本生成图像质量评估模型;利用图像质量评估模型对待评估图像进行评估操作。In a first aspect, an embodiment of the present invention provides an image quality evaluation method, which includes generating an image quality evaluation model using image samples, and performing an evaluation operation on the evaluation image by using an image quality evaluation model.
在本发明一实施例中,利用图像样本生成图像质量评估模型包括标注图像样本的基准质量指标值;根据图像样本生成神经网络结构;利用图像样本对神经网络结构的层级参数进行迭代训练,以生成图像质量评估模型。In an embodiment of the present invention, generating an image quality evaluation model by using an image sample includes labeling a reference quality indicator value of the image sample; generating a neural network structure according to the image sample; and performing iterative training on the hierarchical parameter of the neural network structure by using the image sample to generate Image quality assessment model.
在本发明一实施例中,利用图像样本对神经网络结构的层级参数进行迭代训练,以生成图像质量评估模型包括根据训练参数将图像样本输入至神经网络结构;计算神经网络结构中的损失层的输出结果与基准质量指标值之间的误差数据;利用误差数据更新神经网络结构中的层级参数,并根据更新后的神经网络结构重新迭代计算误差数据;当迭代计算的误差数据达到预设误差范围时,基于神经网络结构生成图像质量评估模型。In an embodiment of the present invention, the image parameters are used to iteratively train the hierarchical parameters of the neural network structure to generate an image quality evaluation model, including inputting the image samples to the neural network structure according to the training parameters; and calculating the loss layer in the neural network structure. The error data between the output result and the reference quality index value; the level parameter in the neural network structure is updated by the error data, and the error data is re-iteratively calculated according to the updated neural network structure; when the error data of the iterative calculation reaches the preset error range An image quality assessment model is generated based on the neural network structure.
在本发明一实施例中,训练参数包括迭代总数、每次迭代样本数目、测试间隔、学习率、初始化神经网络结构的各层级权值、偏置项、偏置项和初始化神经网络结构的各层级权值的学习率中的至少一种。In an embodiment of the invention, the training parameters include the total number of iterations, the number of samples per iteration, the test interval, the learning rate, the initialization of each level of the neural network structure, the bias term, the bias term, and the initialization neural network structure. At least one of the learning rates of the hierarchical weights.
在本发明一实施例中,神经网络结构包括卷积层、激活函数层和损失层。In an embodiment of the invention, the neural network structure includes a convolution layer, an activation function layer, and a loss layer.
在本发明一实施例中,激活函数层包括修正线性单元层和S型生长曲线层。In an embodiment of the invention, the activation function layer includes a modified linear unit layer and an S-type growth curve layer.
在本发明一实施例中,神经网络结构中的层级顺序依次为卷积层、修正线性单元层、卷积层、修正线性单元层、卷积层、修正线性单元层、S型生长曲线层和损失层。In an embodiment of the present invention, the hierarchical order in the neural network structure is a convolution layer, a modified linear unit layer, a convolution layer, a modified linear unit layer, a convolution layer, a modified linear unit layer, an S-type growth curve layer, and Loss layer.
在本发明一实施例中,神经网络结构进一步包括池化层、抛弃层、空间金字塔池化层。In an embodiment of the invention, the neural network structure further includes a pooling layer, a discarding layer, and a spatial pyramid pooling layer.
在本发明一实施例中,神经网络结构中的层级顺序依次为卷积层、修正线性单元层、池化层、卷积层、修正线性单元层、池化层、卷积层、修正线性单元层、池化层、卷积层、修正线性单元层、抛弃层、卷积层、空间金字塔池化层、S型生长曲线层和损失层。In an embodiment of the present invention, the hierarchical order in the neural network structure is a convolution layer, a modified linear unit layer, a pooling layer, a convolution layer, a modified linear unit layer, a pooling layer, a convolution layer, and a modified linear unit. Layer, pooling layer, convolution layer, modified linear unit layer, discarding layer, convolution layer, space pyramid pooling layer, S-type growth curve layer and loss layer.
在本发明一实施例中,卷积层的层级参数包括卷积核数目、卷积核大小、卷积窗口滑动步长及填充边缘像素值。In an embodiment of the invention, the hierarchical parameters of the convolutional layer include a number of convolution kernels, a convolution kernel size, a convolution window sliding step size, and a padding edge pixel value.
在本发明一实施例中,池化层的层级参数包括采样规则、采样窗口大小、采样窗口滑动步长。In an embodiment of the invention, the hierarchical parameters of the pooling layer include a sampling rule, a sampling window size, and a sampling window sliding step size.
在本发明一实施例中,抛弃层的层级参数包括丢弃率。In an embodiment of the invention, the hierarchical parameters of the discard layer include a discard rate.
在本发明一实施例中,空间金字塔池化层的层级参数包括采样规则及金字塔层数。In an embodiment of the invention, the hierarchical parameters of the spatial pyramid pooling layer include sampling rules and a number of pyramid layers.
在本发明一实施例中,图像为文本图像。In an embodiment of the invention, the image is a text image.
第二方面,本发明实施例还提供一种图像质量评估***,该图像质量评估***包括图像质量评估模型生成模块,用于利用图像样本生成图像质量评估模型;评估模块,用于利用图像质量评估模型对待评估图像进行评估操作。In a second aspect, an embodiment of the present invention further provides an image quality evaluation system, where the image quality evaluation system includes an image quality evaluation model generation module for generating an image quality evaluation model using image samples, and an evaluation module for utilizing image quality evaluation The model evaluates the image to be evaluated.
在本发明一实施例中,图像质量评估模型生成模块包括标注单元,用于标注图像样本的基准质量指标值;神经网络结构生成单元,用于根据图像样本生成神经网络结构;迭代训练单元,用于利用图像样本对神经网络结构的层级参数进行迭代训练,以生成图像质量评估模型。In an embodiment of the present invention, the image quality evaluation model generating module includes a labeling unit for labeling a reference quality index value of the image sample; a neural network structure generating unit for generating a neural network structure according to the image sample; and an iterative training unit, Iteratively trains the hierarchical parameters of the neural network structure using image samples to generate an image quality assessment model.
在本发明一实施例中,迭代训练单元还用于根据训练参数将图像样本输入至神经网络结构;计算神经网络结构中的损失层的输出结果与基准质量指标值之间的误差数据;利用误差数据更新神经网络结构中的层级参数,并根据更新后的神经网络结构重新迭代计算误差数据;当迭代计算的误差数据达到预设误差范围时,基于神经网络结构生成图像质量评估模型。In an embodiment of the invention, the iterative training unit is further configured to input the image sample into the neural network structure according to the training parameter; calculate error data between the output result of the loss layer and the reference quality index value in the neural network structure; The data updates the hierarchical parameters in the neural network structure, and recalculates the error data according to the updated neural network structure; when the error data of the iterative calculation reaches the preset error range, the image quality evaluation model is generated based on the neural network structure.
在本发明一实施例中,训练参数包括迭代总数、每次迭代样本数目、测试间隔、学习率、初始化神经网络结构的各层级权值、偏置项、偏置项和初始化神经网络结构的各层级权值的学习率中的至少一种。In an embodiment of the invention, the training parameters include the total number of iterations, the number of samples per iteration, the test interval, the learning rate, the initialization of each level of the neural network structure, the bias term, the bias term, and the initialization neural network structure. At least one of the learning rates of the hierarchical weights.
在本发明一实施例中,神经网络结构包括卷积层、激活函数层和损失层。In an embodiment of the invention, the neural network structure includes a convolution layer, an activation function layer, and a loss layer.
在本发明一实施例中,激活函数层包括修正线性单元层和S型生长曲线层。In an embodiment of the invention, the activation function layer includes a modified linear unit layer and an S-type growth curve layer.
在本发明一实施例中,神经网络结构中的层级顺序依次为卷积层、修正线性单元层、卷积层、修正线性单元层、卷积层、修正线性单元层、S型生长曲线层和损失层。In an embodiment of the present invention, the hierarchical order in the neural network structure is a convolution layer, a modified linear unit layer, a convolution layer, a modified linear unit layer, a convolution layer, a modified linear unit layer, an S-type growth curve layer, and Loss layer.
在本发明一实施例中,神经网络结构进一步包括池化层、抛弃层、空间金字塔池化层。In an embodiment of the invention, the neural network structure further includes a pooling layer, a discarding layer, and a spatial pyramid pooling layer.
在本发明一实施例中,神经网络结构中的层级顺序依次为卷积层、修正线性单元层、池化层、卷积层、修正线性单元层、池化层、卷积层、修正线性单元层、池化层、卷积层、修正线性单元层、抛弃层、卷积层、空间金字塔池化层、S型生长曲线层和损失层。In an embodiment of the present invention, the hierarchical order in the neural network structure is a convolution layer, a modified linear unit layer, a pooling layer, a convolution layer, a modified linear unit layer, a pooling layer, a convolution layer, and a modified linear unit. Layer, pooling layer, convolution layer, modified linear unit layer, discarding layer, convolution layer, space pyramid pooling layer, S-type growth curve layer and loss layer.
在本发明一实施例中,卷积层的层级参数包括卷积核数目、卷积核大小、卷积窗口滑动步长及填充边缘像素值。In an embodiment of the invention, the hierarchical parameters of the convolutional layer include a number of convolution kernels, a convolution kernel size, a convolution window sliding step size, and a padding edge pixel value.
在本发明一实施例中,池化层的层级参数包括采样规则、采样窗口大小、采样窗口滑动步长。In an embodiment of the invention, the hierarchical parameters of the pooling layer include a sampling rule, a sampling window size, and a sampling window sliding step size.
在本发明一实施例中,抛弃层的层级参数包括丢弃率。In an embodiment of the invention, the hierarchical parameters of the discard layer include a discard rate.
在本发明一实施例中,空间金字塔池化层的层级参数包括采样规则及金字塔层数。In an embodiment of the invention, the hierarchical parameters of the spatial pyramid pooling layer include sampling rules and a number of pyramid layers.
在本发明一实施例中,图像为文本图像。In an embodiment of the invention, the image is a text image.
第三方面,本发明实施例还提供一种计算机存储介质,该计算机可读存储介质上存储有图像质量评估程序,该图像质量评估程序被处理器执行时实现上述任一实施例所提及的图像质量评估方法的操作。In a third aspect, an embodiment of the present invention further provides a computer storage medium, where the image quality evaluation program is stored, and when the image quality evaluation program is executed by the processor, the implementation mentioned in any of the above embodiments is implemented. The operation of the image quality assessment method.
本发明实施例提供的图像质量评估方法,利用图像样本生成图像质量评估模型,然后借助 于生成的图像质量评估模型来对待评估图像进行评估操作的方式,完成了对待评估图像的质量评估操作,与现有图像质量评估方法相比,本发明实施例提供的图像质量评估方法计算量低、评估过程简单快捷,提高了评估效率。此外,本发明实施例提供的图像质量评估方法是借助基于图像样本生成的图像质量评估模型来实现评估操作,评估结果具有较高的精准性。尤其是当图像样本和待评估图像的图像类型均为文本图像时,本发明实施例提供的图像质量评估方法能够针对文本图像进行高精准性、高效率的评估操作。The image quality evaluation method provided by the embodiment of the present invention generates an image quality evaluation model by using image samples, and then uses the generated image quality evaluation model to evaluate the image to perform an evaluation operation, and completes the quality evaluation operation of the image to be evaluated, and Compared with the existing image quality evaluation method, the image quality evaluation method provided by the embodiment of the invention has a low calculation amount, a simple and quick evaluation process, and improves the evaluation efficiency. In addition, the image quality evaluation method provided by the embodiment of the present invention implements an evaluation operation by using an image quality evaluation model generated based on image samples, and the evaluation result has high precision. In particular, when the image type of the image sample and the image to be evaluated are both text images, the image quality evaluation method provided by the embodiment of the present invention can perform a highly accurate and highly efficient evaluation operation for the text image.
此外,本发明实施例所提供的图像质量评估***同样具有上述优点和有益效果。In addition, the image quality evaluation system provided by the embodiments of the present invention also has the above advantages and beneficial effects.
附图简要说明BRIEF DESCRIPTION OF THE DRAWINGS
为了更清楚地说明本发明实施例中的技术方案,下面将对实施例描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本发明的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly described below. It is obvious that the drawings in the following description are only some embodiments of the present invention. Other drawings may also be obtained from those of ordinary skill in the art in light of the inventive work.
图1所示为本发明一实施例提供的图像质量评估方法的流程示意图。FIG. 1 is a schematic flowchart diagram of an image quality evaluation method according to an embodiment of the present invention.
图2所示为本发明一实施例提供的图像质量评估方法的利用图像样本生成图像质量评估模型步骤的流程示意图。2 is a flow chart showing the steps of generating an image quality evaluation model using image samples according to an image quality evaluation method according to an embodiment of the present invention.
图3所示为本发明另一实施例提供的神经网络结构的层级结构示意图。FIG. 3 is a schematic diagram showing a hierarchical structure of a neural network structure according to another embodiment of the present invention.
图4所示为本发明又一实施例提供的神经网络结构的层级结构示意图。FIG. 4 is a schematic diagram showing a hierarchical structure of a neural network structure according to another embodiment of the present invention.
图5所示为本发明再一实施例提供的图像质量评估方法的利用图像样本对神经网络结构的层级参数进行迭代训练,以生成图像质量评估模型步骤的流程示意图。FIG. 5 is a flow chart showing the steps of performing an iterative training on the hierarchical parameters of the neural network structure using the image samples to generate an image quality evaluation model according to another embodiment of the present invention.
图6所示为本发明再一实施例提供的迭代训练图像质量评估模型的方法流程图。FIG. 6 is a flowchart of a method for iteratively training an image quality assessment model according to still another embodiment of the present invention.
图7所示为本发明再一实施例提供的用于训练的文本图像样本示例图。FIG. 7 is a diagram showing an example of text image samples for training according to still another embodiment of the present invention.
图8所示为本发明再一实施例提供的文本图像质量评估模型的文本质量网络结构的结构示意图。FIG. 8 is a schematic structural diagram of a text quality network structure of a text image quality assessment model according to still another embodiment of the present invention.
图9所示为本发明再一实施例提供的文本质量网络结构中各层级输出的M*N的灰度图的尺寸变化示意图。FIG. 9 is a schematic diagram showing the size change of the grayscale image of the M*N outputted by each level in the text quality network structure according to another embodiment of the present invention.
图10所示为本发明再一实施例提供的利用训练的文本图像质量评估模型实施文本图像质量评估的流程示意图。FIG. 10 is a flow chart showing the implementation of text image quality assessment using a trained text image quality assessment model according to still another embodiment of the present invention.
图11所示为本发明再一实施例提供的待评估图像的示例图。FIG. 11 is a diagram showing an example of an image to be evaluated according to still another embodiment of the present invention.
图12所示为本发明再一实施例提供的待评估图像的示例图。FIG. 12 is a diagram showing an example of an image to be evaluated according to still another embodiment of the present invention.
图13所示为本发明一实施例提供的图像质量评估***的结构示意图。FIG. 13 is a schematic structural diagram of an image quality evaluation system according to an embodiment of the present invention.
图14所示为本发明另一实施例提供的图像质量评估***的图像质量评估模型生成模块的结构示意图。FIG. 14 is a schematic structural diagram of an image quality evaluation model generating module of an image quality evaluation system according to another embodiment of the present invention.
图15所示为本发明又一实施例提供的文本图像质量评估设备的结构示意图。FIG. 15 is a schematic structural diagram of a text image quality evaluation apparatus according to still another embodiment of the present invention.
图16所示为本发明一实施例提供的电子设备的结构示意图。FIG. 16 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
实施本发明的方式Mode for carrying out the invention
为使本发明的目的、技术方案和优点更加清楚,下面将结合本发明实施例中的附图,对本发明实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本发明一部分实施例,而不是全部的实施例。基于本发明中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本发明保护的范围。The technical solutions in the embodiments of the present invention are clearly and completely described in the following with reference to the accompanying drawings in the embodiments of the present invention. Some embodiments of the invention, rather than all of the embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments of the present invention without creative efforts are within the scope of the present invention.
图1所示为本发明一实施例提供的图像质量评估方法的流程示意图。如图1所示,本发明实施例提供的图像质量评估方法包括:FIG. 1 is a schematic flowchart diagram of an image quality evaluation method according to an embodiment of the present invention. As shown in FIG. 1, the image quality assessment method provided by the embodiment of the present invention includes:
10:利用图像样本生成图像质量评估模型。10: Generate an image quality assessment model using image samples.
图像质量评估模型为根据图像样本训练生成的图像质量评估模型,也就是说,图像质量评估模型中的层级结构以及层级参数等具体信息均根据图像样本的具体情况来训练生成。The image quality evaluation model is an image quality evaluation model generated based on image sample training, that is, specific information such as hierarchical structure and hierarchical parameters in the image quality evaluation model are trained and generated according to the specific conditions of the image samples.
需要说明的是,图像样本的图像类型既可以是包括自然场景的场景图像,又可以是包括文本信息的文本图像,以充分提高本发明实施例提供的图像质量评估方法的适应能力和应用广泛性,本发明实施例对此不进行统一限定。It should be noted that the image type of the image sample may be a scene image including a natural scene or a text image including text information, so as to fully improve the adaptability and application of the image quality evaluation method provided by the embodiment of the present invention. This embodiment of the present invention does not uniformly define this.
优选地,待评估图像的图像类型应当与图像样本的图像类型一致,比如,当图像样本的图像类型为文本图像时,待评估图像的图像类型亦应当为文本图像。应当理解,由于图像质量评估模型的生成依据为图像样本,因此,当待评估图像的图像类型与图像样本的图像类型一致时,能够充分提高图像质量评估模型的评估准确率。Preferably, the image type of the image to be evaluated should be consistent with the image type of the image sample, for example, when the image type of the image sample is a text image, the image type of the image to be evaluated should also be a text image. It should be understood that since the image quality evaluation model is generated based on the image samples, when the image type of the image to be evaluated is consistent with the image type of the image sample, the evaluation accuracy of the image quality evaluation model can be sufficiently improved.
在本发明一实施例中,图像样本和待评估图像的图像类型均为文本图像。In an embodiment of the invention, the image type of the image sample and the image to be evaluated are both text images.
20:利用图像质量评估模型对待评估图像进行评估操作。20: An evaluation operation is performed on the image to be evaluated using the image quality evaluation model.
在实际应用过程中,首先选定图像样本,利用选定的图像样本生成图像质量评估模型,然后再利用生成的图像质量评估模型对待评估图像进行评估操作,以生成待评估图像的评估数据。In the actual application process, the image sample is first selected, the image quality evaluation model is generated by using the selected image sample, and then the evaluation image is evaluated by the generated image quality evaluation model to generate evaluation data of the image to be evaluated.
本发明实施例提供的图像质量评估方法,利用图像样本生成图像质量评估模型,然后借助于生成的图像质量评估模型来对待评估图像进行评估操作的方式,完成了对待评估图像的质量评估操作,与现有图像质量评估方法相比,本发明实施例提供的图像质量评估方法计算量低、评估过程简单快捷,提高了评估效率。此外,本发明实施例提供的图像质量评估方法是借助基于图像样本生成的图像质量评估模型来实现评估操作,评估结果具有较高的精准性。尤其是当图像样本和待评估图像的图像类型均为文本图像时,本发明实施例提供的图像质量评估方法能够针对文本图像进行高精准性、高效率的评估操作。The image quality evaluation method provided by the embodiment of the present invention generates an image quality evaluation model by using image samples, and then uses the generated image quality evaluation model to evaluate the image to perform an evaluation operation, and completes the quality evaluation operation of the image to be evaluated, and Compared with the existing image quality evaluation method, the image quality evaluation method provided by the embodiment of the invention has a low calculation amount, a simple and quick evaluation process, and improves the evaluation efficiency. In addition, the image quality evaluation method provided by the embodiment of the present invention implements an evaluation operation by using an image quality evaluation model generated based on image samples, and the evaluation result has high precision. In particular, when the image type of the image sample and the image to be evaluated are both text images, the image quality evaluation method provided by the embodiment of the present invention can perform a highly accurate and highly efficient evaluation operation for the text image.
图2所示为本发明一实施例提供的图像质量评估方法的利用图像样本生成图像质量评估模型步骤的流程示意图。如图2所示,在本发明实施例提供的图像质量评估方法中,利用图像样本生成图像质量评估模型包括:2 is a flow chart showing the steps of generating an image quality evaluation model using image samples according to an image quality evaluation method according to an embodiment of the present invention. As shown in FIG. 2, in the image quality evaluation method provided by the embodiment of the present invention, generating an image quality evaluation model by using an image sample includes:
11:标注图像样本的基准质量指标值。11: The reference quality indicator value of the image sample.
需要说明的是,图像样本的基准质量指标值的标注,既可以是借助计算机程序等实现的自动标注,又可以是人为手动标注,还可以是其它的标注方式,本发明实施例对此不进行统一限定。It should be noted that the labeling of the reference quality indicator value of the image sample may be an automatic labeling implemented by a computer program or the like, or may be manually labeled manually, or may be other labeling manners, which is not performed by the embodiment of the present invention. Uniform limit.
12:根据图像样本生成神经网络结构。12: Generate a neural network structure from the image samples.
应当理解,神经网络结构为辅助后续机器学习的网络结构,包括输入层、输出层和中间层。It should be understood that the neural network structure is a network structure that assists subsequent machine learning, including an input layer, an output layer, and an intermediate layer.
优选地,神经网络结构为卷积神经网络结构。Preferably, the neural network structure is a convolutional neural network structure.
优选地,神经网络结构包括卷积(Convolution)层、激活函数(Activation Function)层和损失(Loss)层的网络结构。其中,卷积层用于对图像样本进行特征提取操作;激活函数层用于引入非线性因素;损失层用于在训练过程中,根据得出的评估结果和基准质量指标值之间的差异决定是否继续迭代训练。Preferably, the neural network structure includes a network structure of a convolution layer, an activation function layer, and a loss (Loss) layer. The convolution layer is used for feature extraction operations on image samples; the activation function layer is used to introduce nonlinear factors; the loss layer is used in the training process, based on the difference between the obtained evaluation results and the benchmark quality index values. Whether to continue iterative training.
优选地,激活函数层包括修正线性单元(Rectified Linear Unit,ReLU)层和S型生长曲线(Sigmoid)层。Preferably, the activation function layer comprises a Rectified Linear Unit (ReLU) layer and an S-type growth curve (Sigmoid) layer.
更优选地,神经网络结构进一步包括池化(Pooling)层、抛弃(Dropout)层、空间金字塔池化(Spatial Pyramid Pooling,SPP)层。其中,池化层用于进行特征压缩,以简化计算复杂度,降低过拟合;抛弃层用于降低过拟合;空间金字塔池化层用于将提取的特征转换为固定尺寸的特征向量。More preferably, the neural network structure further includes a Pooling layer, a Dropout layer, and a Spatial Pyramid Pooling (SPP) layer. The pooling layer is used for feature compression to simplify computational complexity and reduce over-fitting; the discard layer is used to reduce over-fitting; and the spatial pyramid pooling layer is used to convert the extracted features into fixed-size feature vectors.
13:利用图像样本对神经网络结构的层级参数进行迭代训练,以生成图像质量评估模型。13: Iteratively trains the hierarchical parameters of the neural network structure by using image samples to generate an image quality evaluation model.
优选地,卷积层的层级参数包括卷积核数目、卷积核大小、卷积窗口滑动步长和填充边缘像素值。Preferably, the hierarchical parameters of the convolutional layer include the number of convolution kernels, the convolution kernel size, the convolution window sliding step size, and the padding edge pixel values.
优选地,池化层的层级参数包括采样规则、采样窗口大小和采样窗口滑动步长。Preferably, the hierarchical parameters of the pooled layer include a sampling rule, a sampling window size, and a sampling window sliding step size.
优选地,抛弃层的层级参数包括丢弃率。Preferably, the level parameter of the discard layer comprises a discard rate.
优选地,空间金字塔池化层的层级参数包括采样规则和金字塔层数。Preferably, the hierarchical parameters of the spatial pyramid pooling layer include sampling rules and pyramid layers.
在实际应用过程中,首先选定图像样本,标注图像样本的基准质量指标值,然后根据选定的图像样本生成神经网络结构,并利用图像样本对神经网络结构的层级参数进行迭代训练,以生成图像质量评估模型,最后利用生成的图像质量评估模型对待评估图像进行评估操作,以生成待评估图像的评估数据。In the actual application process, the image sample is first selected, the reference quality index value of the image sample is marked, and then the neural network structure is generated according to the selected image sample, and the image parameters are used to iteratively train the hierarchical parameters of the neural network structure to generate The image quality evaluation model finally uses the generated image quality evaluation model to perform an evaluation operation on the evaluation image to generate evaluation data of the image to be evaluated.
本发明实施例提供的图像质量评估方法,利用图像样本建立基础的神经网络结构,然后基于图像样本对神经网络结构进行深度机器学习,以生成图像质量评估模型的方式,进一步提高了图像质量评估模型的评估精准性。此外,当神经网络结构中包括修正线性单元层、S型生长曲线层、池化层、抛弃层以及空间金字塔池化层等层级结构时,进一步充分提高了所生成的图像质量评估模型的评估精准性和评估效率。The image quality evaluation method provided by the embodiment of the invention further establishes a basic neural network structure by using image samples, and then performs deep machine learning on the neural network structure based on the image samples to generate an image quality evaluation model, thereby further improving the image quality evaluation model. The accuracy of the assessment. In addition, when the neural network structure includes a hierarchical structure such as a modified linear unit layer, an S-type growth curve layer, a pooling layer, an abandoned layer, and a spatial pyramid pooling layer, the evaluation accuracy of the generated image quality evaluation model is further improved. Sex and assessment efficiency.
图3所示为本发明另一实施例提供的神经网络结构的层级结构示意图。如图3所示,在本发明实施例提供的神经网络结构中,层级顺序依次为卷积层、修正线性单元层、卷积层、修正线性单元层、卷积层、修正线性单元层、S型生长曲线层和损失层。FIG. 3 is a schematic diagram showing a hierarchical structure of a neural network structure according to another embodiment of the present invention. As shown in FIG. 3, in the neural network structure provided by the embodiment of the present invention, the hierarchical order is a convolution layer, a modified linear unit layer, a convolution layer, a modified linear unit layer, a convolution layer, a modified linear unit layer, and S. Type growth curve layer and loss layer.
在实际应用过程中,数据的传输顺序依次为卷积层、修正线性单元层、卷积层、修正线性单元层、卷积层、修正线性单元层、S型生长曲线层和损失层,也就是说,数据从位于首层的卷积层输入,最后经损失层输出。In the actual application process, the data transmission order is a convolution layer, a modified linear unit layer, a convolution layer, a modified linear unit layer, a convolution layer, a modified linear unit layer, an S-type growth curve layer, and a loss layer, that is, That is, the data is input from the convolution layer at the first layer and finally output through the loss layer.
图4所示为本发明又一实施例提供的神经网络结构的层级结构示意图。如图4所示,在本发明实施例提供的神经网络结构中,层级顺序依次为卷积层、修正线性单元层、池化层、卷积层、修正线性单元层、池化层、卷积层、修正线性单元层、池化层、卷积层、修正线性单元层、抛弃层、卷积层、空间金字塔池化层、S型生长曲线层和损失层。FIG. 4 is a schematic diagram showing a hierarchical structure of a neural network structure according to another embodiment of the present invention. As shown in FIG. 4, in the neural network structure provided by the embodiment of the present invention, the hierarchical order is a convolution layer, a modified linear unit layer, a pooled layer, a convolution layer, a modified linear unit layer, a pooled layer, and a convolution. Layer, modified linear unit layer, pooled layer, convolutional layer, modified linear unit layer, discarded layer, convolutional layer, spatial pyramid pooling layer, S-type growth curve layer and loss layer.
在实际应用过程中,数据的传输顺序依次为卷积层、修正线性单元层、池化层、卷积层、修正线性单元层、池化层、卷积层、修正线性单元层、池化层、卷积层、修正线性单元层、抛弃层、卷积层、空间金字塔池化层、S型生长曲线层和损失层,也就是说,数据从位于首层的卷积层输入,最后经损失层输出。In the actual application process, the data transmission order is convolution layer, modified linear unit layer, pooling layer, convolution layer, modified linear unit layer, pooling layer, convolution layer, modified linear unit layer, and pooling layer. , convolutional layer, modified linear unit layer, discarding layer, convolution layer, spatial pyramid pooling layer, S-type growth curve layer and loss layer, that is, data is input from the convolution layer located in the first layer, and finally lost Layer output.
图5所示为本发明再一实施例提供的图像质量评估方法的利用图像样本对神经网络结构的层级参数进行迭代训练,以生成图像质量评估模型步骤的流程示意图。如图5所示,在本发明实施例提供的图像质量评估方法中,利用图像样本对神经网络结构的层级参数进行迭代训练,以生成图像质量评估模型包括:FIG. 5 is a flow chart showing the steps of performing an iterative training on the hierarchical parameters of the neural network structure using the image samples to generate an image quality evaluation model according to another embodiment of the present invention. As shown in FIG. 5, in the image quality evaluation method provided by the embodiment of the present invention, iteratively training the hierarchical parameters of the neural network structure by using image samples to generate an image quality evaluation model includes:
131:根据训练参数将图像样本输入至神经网络结构。131: Input an image sample to the neural network structure according to the training parameter.
优选地,训练参数包括迭代总数、每次迭代样本数目、测试间隔、学习率、初始化神经网络结构的各层级权值、偏置项、偏置项和初始化神经网络结构的各层级权值的学习率中的至少一种。Preferably, the training parameters include the total number of iterations, the number of samples per iteration, the test interval, the learning rate, the initialization of the hierarchical weights of the neural network structure, the bias term, the bias term, and the learning of the hierarchical weights of the neural network structure. At least one of the rates.
132:计算神经网络结构中的损失层的输出结果与基准质量指标值之间的误差数据。132: Calculate error data between the output result of the loss layer in the neural network structure and the reference quality indicator value.
133:利用误差数据更新神经网络结构中的层级参数,并根据更新后的神经网络结构重新迭代计算误差数据。133: Update the hierarchical parameters in the neural network structure by using the error data, and re-iteratively calculate the error data according to the updated neural network structure.
134:当迭代计算的误差数据达到预设误差范围时,基于神经网络结构生成图像质量评估模型。134: When the error data of the iterative calculation reaches a preset error range, an image quality evaluation model is generated based on the neural network structure.
需要说明的是,预设误差范围可根据实际情况自行设定,以充分提高本发明实施例提供的图像质量评估方法的适应能力和应用广泛性,本发明实施例对此不进行统一限定。It should be noted that the preset error range can be set according to the actual situation, so as to fully improve the adaptability and application of the image quality evaluation method provided by the embodiment of the present invention, which is not limited by the embodiment of the present invention.
在实际应用过程中,首先选定图像样本,标注图像样本的基准质量指标值,并根据选定的图像样本生成神经网络结构,然后选定训练参数,根据训练参数将图像样本输入至生成的神经网络结构,计算神经网络结构中的损失层的输出结果与基准质量指标值之间的误差数据,利用误差数据更新神经网络结构中的层级参数,并根据更新后的神经网络结构重新迭代计算误差数据,当迭代计算的误差数据达到预设误差范围时,该误差数据对应的神经网络结构即为图像质量评估模型,最后利用生成的图像质量评估模型对待评估图像进行评估操作,以生成待评估图像的评估数据。In the actual application process, the image sample is first selected, the reference quality index value of the image sample is marked, and the neural network structure is generated according to the selected image sample, then the training parameter is selected, and the image sample is input to the generated nerve according to the training parameter. The network structure calculates the error data between the output result of the loss layer and the reference quality index value in the neural network structure, uses the error data to update the hierarchical parameters in the neural network structure, and re-iteratively calculates the error data according to the updated neural network structure. When the error data of the iterative calculation reaches the preset error range, the neural network structure corresponding to the error data is an image quality evaluation model, and finally the image quality evaluation model is used to perform an evaluation operation on the image to be evaluated to generate an image to be evaluated. Evaluation data.
图6所示为本发明再一实施例提供的迭代训练图像质量评估模型的方法流程图。在本发明实施例中,图像样本和待评估图像的图像类型均为文本图像。FIG. 6 is a flowchart of a method for iteratively training an image quality assessment model according to still another embodiment of the present invention. In an embodiment of the invention, the image type of the image sample and the image to be evaluated are both text images.
在本发明实施例中,需要预先训练文本图像质量评估模型,再通过该文本图像质量评估模型对文本图像(即待评估图像)的质量进行评估。In the embodiment of the present invention, the text image quality assessment model needs to be pre-trained, and the quality of the text image (ie, the image to be evaluated) is evaluated by the text image quality assessment model.
参照图6所示,该训练文本图像质量评估模型的过程具体包括:Referring to FIG. 6, the process of the training text image quality assessment model specifically includes:
101:获取训练的文本图像样本,并对每个文本图像样本进行质量指标值标注。101: Obtain a sample of the trained text image, and perform quality indicator value labeling on each text image sample.
应当理解,101中提及的质量指标值即为上述实施例中所提及的基准质量指标值。It should be understood that the quality index value mentioned in 101 is the reference quality index value mentioned in the above embodiment.
具体地,该用于训练的文本图像样本包括从公开的文本图像质量数据库中获取的文本图像样本,还包括合成的文本图像样本;该文本图像样本可以为多光谱图像、普通彩色图像或灰度图像;文本图像样本中的文本包括中文、英文以及其他语音字符的文本图像。Specifically, the text image sample for training includes a text image sample obtained from a public text image quality database, and further includes a synthesized text image sample; the text image sample may be a multi-spectral image, a normal color image, or a grayscale Image; text in a sample of text images includes text images of Chinese, English, and other phonetic characters.
对每个文本图像样本进行质量指标值标注包括:可以采用计算机自动标注,比如将OCR识别的文本图像的质量参数作为文本图像的质量指标值;也可以采用人工手动标注,通过人眼观察文本图像,并标注文本图像的质量指标值;还可以采用其他方式对文本图像样本进行质量指标值标注,本发明实施例对采用的具体方式不加以限定。The quality indicator value labeling for each text image sample includes: automatic labeling by computer, for example, the quality parameter of the text image recognized by the OCR as the quality index value of the text image; or manual manual labeling, the text image is observed by the human eye And the quality index value of the text image is marked; the quality index value is also marked on the text image sample in other manners, and the specific manner adopted in the embodiment of the present invention is not limited.
其中,文本图像的质量指标值可以用浮点数表示,浮点数越大,表示文本图像质量越好;对每个文本图像样本进行质量指标值标注后,还包括将所有标注的文本图像的质量指标值的大小缩放到浮点数0-0.1的范围内。The quality index value of the text image can be represented by a floating point number. The larger the floating point number, the better the quality of the text image; after the quality index value is marked for each text image sample, the quality index of all the labeled text images is also included. The size of the value is scaled to a range of 0-0.1 for the floating point number.
示例性的,图7所示为本发明再一实施例提供的用于训练的文本图像样本示例图。如图7所示,图7中的a、b、c、d四个文本图像可以作为训练文本图像质量评估模型使用的文本图像样本,其中,对该a、b、c、d四个文本图像样本标注的质量指标值分别为:0.91,0.8658,0.2733,0.9067。Illustratively, FIG. 7 is a diagram showing an example of text image samples for training according to still another embodiment of the present invention. As shown in FIG. 7, four text images a, b, c, and d in FIG. 7 can be used as text image samples used in the training text image quality evaluation model, wherein four text images of a, b, c, and d are used. The quality index values of the sample labels are: 0.91, 0.8658, 0.2733, 0.9067.
102:设置文本图像质量评估模型的文本质量网络。102: Set the text quality network of the text image quality assessment model.
需要说明的是,该文本图像质量评估模型是基于深度学习的卷积神经网络进行训练的,所以需要设置神经网络结构,即文本质量网络。It should be noted that the text image quality assessment model is trained based on a deep learning convolutional neural network, so a neural network structure, that is, a text quality network, needs to be set.
首先,建立文本质量网络结构的基础层级结构。具体地,文本质量网络结构中包括五个Convolution层、四个ReLU层、三个Pooling层、一个Dropout层、一个SPP层、一个Sigmoid层、一个Loss层,并且,文本质量网络结构的层级排布顺序为:Convolution/ReLU/Pooling/Convolution/ReLU/Pooling/Convolution/ReLU/Pooling/Convolution/ReLU/Dropout/Convolution/SPP/Sigmoid/Loss。First, establish the basic hierarchical structure of the text quality network structure. Specifically, the text quality network structure includes five Convolution layers, four ReLU layers, three Pooling layers, one Dropout layer, one SPP layer, one Sigmoid layer, one Loss layer, and a hierarchical arrangement of text quality network structures. The order is: Convolution/ReLU/Pooling/Convolution/ReLU/Pooling/Convolution/ReLU/Pooling/Convolution/ReLU/Dropout/Convolution/SPP/Sigmoid/Loss.
示例性的,图8所示为本发明再一实施例提供的文本图像质量评估模型的文本质量网络结构的结构示意图。具体地,图8所示为文本质量网络结构中的层级排布的结构示意图,其中,CONV即Convolution,POOL即Pooling,DROP即Dropout。Illustratively, FIG. 8 is a schematic structural diagram of a text quality network structure of a text image quality assessment model according to still another embodiment of the present invention. Specifically, FIG. 8 is a schematic structural diagram of a hierarchical arrangement in a text quality network structure, where CONV is Convolution, POOL is Pooling, and DROP is Dropout.
然后,设置文本质量网络结构中Convolution层、Pooling层、Dropout层及SPP层的层级参数。Then, set the hierarchy parameters of the Convolution layer, the Pooling layer, the Dropout layer, and the SPP layer in the text quality network structure.
其中,设置的Convolution层的层级参数包括卷积核数目、卷积核大小、卷积窗口滑动步长及填充边缘像素值;Pooling层的层级参数包括采样规则、采样窗口大小、采样窗口滑动步长; Dropout层的层级参数包括丢弃率;SPP层的层级参数包括采样规则及金字塔层数。The hierarchical parameters of the set Convolution layer include the number of convolution kernels, the convolution kernel size, the convolution window sliding step size, and the padding edge pixel values; the layering parameters of the Pooling layer include sampling rules, sampling window size, sampling window sliding step size The hierarchical parameters of the Dropout layer include the discard rate; the hierarchical parameters of the SPP layer include the sampling rules and the number of pyramid layers.
优选地,Convolution层、Pooling层、Dropout层和SPP层的层级参数设置如下:第一Convolution层的卷积核数目为96,卷积核大小为3*3,卷积窗口滑动步长为1,填充边缘像素值为0;第一Pooling层的采样规则为最大值采样,采样窗口大小为3,采样窗口滑动步长为2;第二Convolution层的卷积核数目为96,卷积核大小为3*3,卷积窗口滑动步长为1,填充边缘像素值为2;第二Pooling层的采样规则为最大值采样,采样窗口大小为3*3,采样窗口滑动步长为2;第三Convolution层的卷积核数目为128,卷积核大小为3*3,卷积窗口滑动步长为1,填充边缘像素值为1;第三层Pooling层的采样规则为最大值采样,采样窗口大小为3,采样窗口滑动步长为2;第四Convolution层的卷积核数目为192,卷积核大小为1,卷积窗口滑动步长为1,填充边缘像素值为0;Dropout层丢弃率为0.35;第五Convolution层的卷积核数目为1,卷积核大小为1,卷积窗口滑动步长为1,填充边缘像素为0;SPP层采样规则为最大值采样,金字塔层数为1。Preferably, the hierarchical parameters of the Convolution layer, the Pooling layer, the Dropout layer, and the SPP layer are set as follows: the first Convolution layer has a number of convolution kernels of 96, the convolution kernel size is 3*3, and the convolution window has a sliding step size of 1. The padding edge pixel value is 0; the sampling rule of the first Pooling layer is the maximum value sampling, the sampling window size is 3, the sampling window sliding step size is 2; the second Convolution layer has a convolution kernel number of 96, and the convolution kernel size is 3*3, the convolution window has a sliding step size of 1, and the padding edge pixel value is 2; the sampling rule of the second Pooling layer is the maximum value sampling, the sampling window size is 3*3, and the sampling window sliding step size is 2; The convolution layer has a convolution kernel number of 128, a convolution kernel size of 3*3, a convolution window sliding step size of 1, and a padding edge pixel value of 1; a sampling rule of the third layer of the Pooling layer is a maximum value sampling, sampling window The size is 3, the sampling window sliding step size is 2; the fourth Convolution layer has a convolution kernel number of 192, the convolution kernel size is 1, the convolution window sliding step size is 1, the padding edge pixel value is 0; and the Dropout layer discards The rate is 0.35; the convolution kernel of the fifth Convolution layer The size is 1, the convolution kernel size is 1, the convolution window sliding step size is 1, the padding edge pixel is 0; the SPP layer sampling rule is the maximum value sampling, and the pyramid layer number is 1.
需要说明的是,以上所示的Convolution层、Pooling层、Dropout层和SPP层的层级参数的设置值为优选值,在实际训练过程中根据需要可进行调整,本发明实施例对各层级的具体层级参数不加以限定。It should be noted that the setting values of the tier parameters of the Convolution layer, the Pooling layer, the Dropout layer, and the SPP layer are the preferred values, and can be adjusted as needed during the actual training process, and the embodiments of the present invention are specific to each level. Hierarchy parameters are not limited.
103:基于文本图像样本及标注的质量指标值,通过文本质量网络对初始的文本图像质量评估模型的参数进行迭代训练,以获取文本图像质量评估模型。103: Perform an iterative training on parameters of the initial text image quality assessment model based on the text image sample and the labeled quality indicator value to obtain a text image quality assessment model.
具体地,103中所提及的获取文本图像质量评估模型的过程包括:Specifically, the process of obtaining a text image quality assessment model mentioned in 103 includes:
a、确定训练参数。a. Determine the training parameters.
具体地,训练参数包括迭代总数、每次迭代样本数目、测试间隔、学习率、初始化网络各层权值、偏置项、以及偏置项和初始化网络各层权值的学习率中的至少一种。Specifically, the training parameters include at least one of a total number of iterations, a number of samples per iteration, a test interval, a learning rate, an initialization network layer weight, an offset term, and a learning rate of the offset term and the initialization network layer weights. Kind.
b、根据训练参数,将文本图像样本输入初始的文本图像质量评估模型。b. Input the text image sample into the initial text image quality assessment model according to the training parameters.
c、获取文本质量网络的Convolution层、ReLU层、Pooling层、Dropout层、SPP层、Sigmoid层对文本图像样本进行处理的输出结果。c. Obtain an output result of processing the text image sample by the Convolution layer, the ReLU layer, the Pooling layer, the Dropout layer, the SPP layer, and the Sigmoid layer of the text quality network.
具体地,训练的文本图像样本数据依次经过各个Convolution层、ReLU层、Pooling层的操作,以及经过Dropout层、SPP层和Sigmoid层的网络运算,并在Sigmoid层获取最终的输出结果。Specifically, the trained text image sample data sequentially passes through operations of respective Convolution layers, ReLU layers, and Pooling layers, and network operations through the Dropout layer, the SPP layer, and the Sigmoid layer, and obtains final output results in the Sigmoid layer.
d、在文本质量网络的Loss层计算输出结果与标注的质量指标值之间的误差,并将误差反向传播到文本质量网络结构各层,以更新各层的层级参数,迭代计算直至误差达到预设范围。d. Calculate the error between the output result and the labeled quality index value in the Loss layer of the text quality network, and propagate the error back to the layers of the text quality network structure to update the hierarchical parameters of each layer, and iteratively calculate until the error reaches Preset range.
具体的,更新各层的层级参数包括更新初始化网络各层的权值。Specifically, updating the hierarchical parameters of each layer includes updating the weights of each layer of the initialization network.
图9所示为本发明再一实施例提供的文本质量网络结构中各层级输出的M*N的灰度图的尺寸变化示意图。进一步地,为了说明该训练过程,以对尺寸为M*N的灰度图像为例进行说明,将该M*N的灰度图像输入训练中的文本图像质量评估模型中,在该文本图像质量评估模型的文本质量网络结构各层输出的尺寸变化可以参照图9所示,具体如下:FIG. 9 is a schematic diagram showing the size change of the grayscale image of the M*N outputted by each level in the text quality network structure according to another embodiment of the present invention. Further, in order to illustrate the training process, a grayscale image of size M*N is taken as an example for description, and the grayscale image of the M*N is input into the text image quality evaluation model in training, in which the text image quality is The text quality of the evaluation model The size change of the output of each layer of the network structure can be referred to Figure 9, as follows:
第一层Convolution层返回的结构大小为1×96×M×N;The structure returned by the first layer of Convolution layer is 1×96×M×N;
第一层Pooling层返回的结构大小为1×96×M/2×N/2;The structure returned by the first layer of the Pooling layer is 1×96×M/2×N/2;
第二层Convolution层返回的结构大小为1×96×M/2×N/2;The structure returned by the second layer of Convolution layer is 1×96×M/2×N/2;
第二层Pooling层返回的结构大小为1×96×M/4×N/4;The structure returned by the second layer of the Pooling layer is 1×96×M/4×N/4;
第三层Convolution层返回的结构大小为1×128×M/4×N/4;The size of the structure returned by the third layer of Convolution layer is 1×128×M/4×N/4;
第三层Pooling层返回的结构大小为1×128×M/8×N/8;The structure returned by the third layer of the Pooling layer is 1×128×M/8×N/8;
第四层Convolution层返回的结构大小为1×192×M/8×N/8;The structure size returned by the fourth layer Convolution layer is 1×192×M/8×N/8;
第五层Convolution层返回的结构大小为1×1×M/8×N/8;The size of the structure returned by the fifth layer Convolution layer is 1×1×M/8×N/8;
SPP层返回的结构大小为1×1×1×1;The structure returned by the SPP layer is 1×1×1×1;
Sigmoid层返回的结构大小为1×1×1×1;The size of the structure returned by the Sigmoid layer is 1×1×1×1;
再将Sigmoid层的输出结果在Loss层与该文本图像样本标注的质量指标值进行误差运算,并将误差返回以上各层,更新各层的网络参数。使用大量文本图像样本迭代该过程直至Loss层的误差达到预设范围。Then, the output result of the Sigmoid layer is subjected to error calculation in the Loss layer and the quality index value of the text image sample, and the error is returned to the above layers, and the network parameters of each layer are updated. The process is iterated using a large number of text image samples until the error of the Loss layer reaches a preset range.
e、获取最终生成的文本图像质量评估模型。e. Obtain a final generated text image quality assessment model.
图10所示为本发明再一实施例提供的利用训练的文本图像质量评估模型实施文本图像质量评估的流程示意图。参照图10所示,利用训练生成的文本图像质量评估模型进行文本图像质量评估的过程,具体包括:FIG. 10 is a flow chart showing the implementation of text image quality assessment using a trained text image quality assessment model according to still another embodiment of the present invention. Referring to FIG. 10, a process for text image quality assessment using a text image quality assessment model generated by training includes:
501:获取所需评估的文本图像。501: Obtain a text image of the desired evaluation.
具体地,该所需评估的文本图像与图6所示的101中的文本图像样本类似,此处不再加以赘述,101中所述文本图像样本的特征均可以用于此处的文本图像。Specifically, the text image of the desired evaluation is similar to the text image sample in 101 shown in FIG. 6, and details are not described herein. The features of the text image sample in 101 can be used for the text image herein.
502:将文本图像输入预先训练的文本图像质量评估模型中处理,根据文本图像质量评估模型的输出值,确定文本图像的质量指标值。502: Input the text image into the pre-trained text image quality assessment model, and determine the quality index value of the text image according to the output value of the text image quality evaluation model.
该质量指标值包括浮点数。The quality indicator value includes a floating point number.
示例性的,图11所示为本发明再一实施例提供的待评估图像的示例图。图12所示为本发明再一实施例提供的待评估图像的示例图。将图11和图12中所示的文本图像输出该预先训练的该文本图像质量评估模型进行处理,经处理获取的图11中所示的文本图像的质量指标值为0.9756,获取的图12中所示的文本图像的质量指标值为0.9805。Illustratively, FIG. 11 is a diagram showing an example of an image to be evaluated according to still another embodiment of the present invention. FIG. 12 is a diagram showing an example of an image to be evaluated according to still another embodiment of the present invention. The text image shown in FIG. 11 and FIG. 12 is outputted to the pre-trained text image quality evaluation model for processing, and the quality index value of the text image shown in FIG. 11 obtained by the processing is 0.9756, which is obtained in FIG. The quality index value of the text image shown is 0.9805.
本发明实施例提供了一种文本图像质量评估方法,通过获取所需评估的文本图像,将该文本图像输入预先训练的文本图像质量评估模型中,根据文本图像质量评估模型的输出值,确定文本图像的质量指标值,从而通过预先训练的文本图像质量评估模型,能够专门用于对文本图像的质量进行评估,且该评估过程简便易于操作,可以作为OCR前的预处理操作,能够减少计算消耗,与现有技术中的图像质量评估方法相比,大大降低了计算复杂度和计算量,并且与现有技术相比无需对图像进行预处理等操作,评估过程快捷;另外,因为该文本图像质量评估模型是基于深度学习的神经网络进行训练生成,在评估过程中能够模拟人类视觉对文本图像质量的评估过程,且该模型的参数通过迭代进行反复训练,所以通过该预先训练的文本图像质量评估模型对文本图像的质量进行评估,能够提供更加有效和更精准的质量评估结果,提高了评估效率。An embodiment of the present invention provides a text image quality evaluation method, which obtains a text image to be evaluated, inputs the text image into a pre-trained text image quality evaluation model, and determines a text according to an output value of the text image quality evaluation model. The quality index value of the image, which can be used to evaluate the quality of the text image through the pre-trained text image quality evaluation model, and the evaluation process is simple and easy to operate, and can be used as a pre-processing operation before OCR, which can reduce computational consumption. Compared with the image quality assessment method in the prior art, the calculation complexity and the calculation amount are greatly reduced, and the image is not required to be pre-processed compared with the prior art, and the evaluation process is fast; in addition, because the text image The quality assessment model is based on the deep learning neural network for training generation. In the evaluation process, the human visual image quality evaluation process can be simulated, and the parameters of the model are iteratively trained repeatedly, so the quality of the pre-trained text image is adopted. Evaluation model for text images Quality assessment, to provide more efficient and more accurate quality assessment, assessment of improved efficiency.
图13所示为本发明一实施例提供的图像质量评估***的结构示意图。如图13所示,本发明实施例提供的图像质量评估***包括:FIG. 13 is a schematic structural diagram of an image quality evaluation system according to an embodiment of the present invention. As shown in FIG. 13, the image quality evaluation system provided by the embodiment of the present invention includes:
图像质量评估模型生成模块100,用于利用图像样本生成图像质量评估模型。The image quality assessment model generation module 100 is configured to generate an image quality assessment model using the image samples.
评估模块200,用于利用图像质量评估模型对待评估图像进行评估操作。The evaluation module 200 is configured to perform an evaluation operation on the image to be evaluated by using the image quality evaluation model.
注意,图像质量评估模型为根据图像样本训练生成的图像质量评估模型,也就是说,图像质量评估模型中的层级结构以及层级参数等具体信息均根据图像样本的具体情况来训练生成。Note that the image quality evaluation model is an image quality evaluation model generated based on image sample training, that is, specific information such as hierarchical structure and hierarchical parameters in the image quality evaluation model are trained and generated according to the specific conditions of the image samples.
需要说明的是,图像样本的图像类型既可以是包括自然场景的场景图像,又可以是包括文本信息的文本图像,以充分提高本发明实施例提供的图像质量评估方法的适应能力和应用广泛性,本发明实施例对此不进行统一限定。It should be noted that the image type of the image sample may be a scene image including a natural scene or a text image including text information, so as to fully improve the adaptability and application of the image quality evaluation method provided by the embodiment of the present invention. This embodiment of the present invention does not uniformly define this.
优选地,待评估图像的图像类型应当与图像样本的图像类型一致,比如,当图像样本的图像类型为文本图像时,待评估图像的图像类型亦应当为文本图像。应当理解,由于图像质量评估模型的生成依据为图像样本,因此,当待评估图像的图像类型与图像样本的图像类型一致时,能够充分提高图像质量评估模型的评估准确率。Preferably, the image type of the image to be evaluated should be consistent with the image type of the image sample, for example, when the image type of the image sample is a text image, the image type of the image to be evaluated should also be a text image. It should be understood that since the image quality evaluation model is generated based on the image samples, when the image type of the image to be evaluated is consistent with the image type of the image sample, the evaluation accuracy of the image quality evaluation model can be sufficiently improved.
优选地,图像样本和待评估图像的图像类型均为文本图像。Preferably, the image type and the image type of the image to be evaluated are both text images.
图14所示为本发明另一实施例提供的图像质量评估***的图像质量评估模型生成模块的 结构示意图。如图14所示,在本发明实施例中,图像质量评估模型生成模块100包括:FIG. 14 is a schematic structural diagram of an image quality evaluation model generating module of an image quality evaluation system according to another embodiment of the present invention. As shown in FIG. 14, in the embodiment of the present invention, the image quality assessment model generation module 100 includes:
标注单元110,用于标注图像样本的基准质量指标值。The labeling unit 110 is configured to label a reference quality indicator value of the image sample.
神经网络结构生成单元120,用于根据图像样本生成神经网络结构。The neural network structure generating unit 120 is configured to generate a neural network structure according to the image samples.
迭代训练单元130,用于利用图像样本对神经网络结构的层级参数进行迭代训练,以生成图像质量评估模型。The iterative training unit 130 is configured to perform iterative training on the hierarchical parameters of the neural network structure by using image samples to generate an image quality evaluation model.
应当理解,神经网络结构为辅助后续机器学习的网络结构,包括输入层、输出层和中间层。It should be understood that the neural network structure is a network structure that assists subsequent machine learning, including an input layer, an output layer, and an intermediate layer.
优选地,神经网络结构为卷积神经网络结构。Preferably, the neural network structure is a convolutional neural network structure.
优选地,神经网络结构包括卷积(Convolution)层、激活函数(Activation Function)层和损失(Loss)层的网络结构。其中,卷积层用于对图像样本进行特征提取操作;激活函数层用于引入非线性因素;损失层用于在训练过程中,根据得出的评估结果和基准质量指标值之间的差异决定是否继续迭代训练。Preferably, the neural network structure includes a network structure of a convolution layer, an activation function layer, and a loss (Loss) layer. The convolution layer is used for feature extraction operations on image samples; the activation function layer is used to introduce nonlinear factors; the loss layer is used in the training process, based on the difference between the obtained evaluation results and the benchmark quality index values. Whether to continue iterative training.
优选地,激活函数层包括修正线性单元(Rectified Linear Unit,ReLU)层和S型生长曲线(Sigmoid)层。Preferably, the activation function layer comprises a Rectified Linear Unit (ReLU) layer and an S-type growth curve (Sigmoid) layer.
更优选地,神经网络结构进一步包括池化(Pooling)层、抛弃(Dropout)层、空间金字塔池化(Spatial Pyramid Pooling,SPP)层。其中,池化层用于进行特征压缩,以简化计算复杂度,降低过拟合;抛弃层用于降低过拟合;空间金字塔池化层用于将提取的特征转换为固定尺寸的特征向量。More preferably, the neural network structure further includes a Pooling layer, a Dropout layer, and a Spatial Pyramid Pooling (SPP) layer. The pooling layer is used for feature compression to simplify computational complexity and reduce over-fitting; the discard layer is used to reduce over-fitting; and the spatial pyramid pooling layer is used to convert the extracted features into fixed-size feature vectors.
优选地,卷积层的层级参数包括卷积核数目、卷积核大小、卷积窗口滑动步长和填充边缘像素值。Preferably, the hierarchical parameters of the convolutional layer include the number of convolution kernels, the convolution kernel size, the convolution window sliding step size, and the padding edge pixel values.
优选地,池化层的层级参数包括采样规则、采样窗口大小和采样窗口滑动步长。Preferably, the hierarchical parameters of the pooled layer include a sampling rule, a sampling window size, and a sampling window sliding step size.
优选地,抛弃层的层级参数包括丢弃率。Preferably, the level parameter of the discard layer comprises a discard rate.
优选地,空间金字塔池化层的层级参数包括采样规则和金字塔层数。Preferably, the hierarchical parameters of the spatial pyramid pooling layer include sampling rules and pyramid layers.
在本发明另一实施例中,迭代训练单元130还用于根据训练参数将图像样本输入至神经网络结构,计算神经网络结构中的损失层的输出结果与基准质量指标值之间的误差数据,利用误差数据更新神经网络结构中的层级参数,并根据更新后的神经网络结构重新迭代计算误差数据,当迭代计算的误差数据达到预设误差范围时,误差数据对应的神经网络结构即为图像质量评估模型。In another embodiment of the present invention, the iterative training unit 130 is further configured to input an image sample into the neural network structure according to the training parameter, and calculate error data between the output result of the loss layer and the reference quality indicator value in the neural network structure. The error parameters are used to update the hierarchical parameters in the neural network structure, and the error data is re-iteratively calculated according to the updated neural network structure. When the error data of the iterative calculation reaches the preset error range, the neural network structure corresponding to the error data is the image quality. Evaluation model.
应当理解,图13和图14提供的图像质量评估***中的图像质量评估模型生成模块100、评估模块200、标注单元110、神经网络结构生成单元120和迭代训练单元130的操作和功能可以参考上述图1、图2和图5提供的图像质量评估方法,为了避免重复,在此不再赘述。It should be understood that the operations and functions of the image quality evaluation model generating module 100, the evaluation module 200, the labeling unit 110, the neural network structure generating unit 120, and the iterative training unit 130 in the image quality evaluation system provided in FIG. 13 and FIG. The image quality evaluation methods provided in FIG. 1, FIG. 2 and FIG. 5 are not described herein again in order to avoid repetition.
图15所示为本发明又一实施例提供的文本图像质量评估设备的结构示意图。参照图15所示,本发明实施例提供的文本图像质量评估设备8包括:FIG. 15 is a schematic structural diagram of a text image quality evaluation apparatus according to still another embodiment of the present invention. Referring to FIG. 15, the text image quality evaluating apparatus 8 provided by the embodiment of the present invention includes:
文本图像获取模块81,用于获取所需评估的文本图像;a text image obtaining module 81, configured to acquire a text image to be evaluated;
评估模块82,用于将文本图像输入预先训练的文本图像质量评估模型中处理,根据文本图像质量评估模型的输出值,确定文本图像的质量指标值,质量指标值包括浮点数。The evaluation module 82 is configured to input the text image into the pre-trained text image quality assessment model, and determine a quality index value of the text image according to the output value of the text image quality evaluation model, where the quality indicator value includes a floating point number.
进一步地,文本图像质量评估设备8还包括:Further, the text image quality evaluation device 8 further includes:
文本图像样本获取模块83,用于获取训练的文本图像样本;a text image sample obtaining module 83, configured to acquire a trained text image sample;
质量指标值标注模块84,用于对每个文本图像样本进行质量指标值标注;a quality indicator value labeling module 84, configured to perform quality indicator value labeling on each text image sample;
文本质量网络设置模块85,用于设置文本图像质量评估模型的文本质量网络;a text quality network setting module 85, configured to set a text quality network of the text image quality assessment model;
文本图像质量评估模型获取模块86,用于基于文本图像样本及标注的质量指标值,通过文本质量网络对初始的文本图像质量评估模型的参数进行迭代计算训练,以获取文本图像质量评估模型。The text image quality evaluation model obtaining module 86 is configured to perform iterative calculation training on the parameters of the initial text image quality evaluation model through the text quality network based on the text image sample and the labeled quality index value to obtain the text image quality evaluation model.
进一步地,文本质量网络设置模块85包括:Further, the text quality network setting module 85 includes:
文本质量网络结构设置子模块851,用于设置文本质量网络结构,文本质量网络结构由五个Convolution层、四个ReLU层、三个Pooling层、一个Dropout层、一个SPP层、一个Sigmoid层、一个Loss层构成;The text quality network structure setting sub-module 851 is configured to set a text quality network structure. The text quality network structure consists of five Convolution layers, four ReLU layers, three Pooling layers, one Dropout layer, one SPP layer, one Sigmoid layer, and one Composition of the Loss layer;
文本质量网络结构的顺序为:Convolution/ReLU/Pooling/Convolution/ReLU/Pooling/Convolution/ReLU/Pooling/Convolution/ReLU/Dropout/Convolution/SPP/Sigmoid/Loss;以及The order of the text quality network structure is: Convolution/ReLU/Pooling/Convolution/ReLU/Pooling/Convolution/ReLU/Pooling/Convolution/ReLU/Dropout/Convolution/SPP/Sigmoid/Loss;
层结构参数设置子模块852,用于设置文本质量网络结构Convolution层、Pooling层、Dropout层及SPP层的层级参数。The layer structure parameter setting sub-module 852 is configured to set hierarchical parameters of the text quality network structure Convolution layer, the Pooling layer, the Dropout layer, and the SPP layer.
进一步地,设置的Convolution层的层级参数包括卷积核数目、卷积核大小、卷积窗口滑动步长及填充边缘像素值;Pooling层的层级参数包括采样规则、采样窗口大小、采样窗口滑动步长;Dropout层的层级参数包括丢弃率;SPP层的层级参数包括采样规则及金字塔层数。Further, the hierarchical parameters of the set Convolution layer include a number of convolution kernels, a convolution kernel size, a convolution window sliding step size, and a padding edge pixel value; the hierarchical parameters of the Pooling layer include a sampling rule, a sampling window size, and a sampling window sliding step. Long; the hierarchical parameters of the Dropout layer include the discard rate; the hierarchical parameters of the SPP layer include the sampling rules and the number of pyramid layers.
进一步地,文本图像质量评估模型获取模块86具体包括:Further, the text image quality assessment model obtaining module 86 specifically includes:
训练参数确定子模块861,确定训练参数;Training parameter determination sub-module 861, determining training parameters;
输入子模块862,用于根据训练参数,将文本图像样本输入初始的文本图像质量评估模型;The input sub-module 862 is configured to input the text image sample into the initial text image quality assessment model according to the training parameter;
输出结果获取子模块863,用于获取文本质量网络的Convolution层、ReLU层、Pooling层、Dropout层、SPP层、Sigmoid层对文本图像样本进行处理的输出结果;The output result obtaining submodule 863 is configured to obtain an output result of processing a text image sample by a Convolution layer, a ReLU layer, a Pooling layer, a Dropout layer, an SPP layer, and a Sigmoid layer of the text quality network;
误差计算子模块864,在文本质量网络的Loss层计算输出结果与标注的质量指标值之间的误差;The error calculation sub-module 864 calculates an error between the output result and the labeled quality indicator value in the Loss layer of the text quality network;
迭代计算模块865,用于将误差反向传播到文本质量网络结构各层,以更新各层的网络参数,迭代计算直至误差达到预设范围;An iterative calculation module 865 is configured to inversely propagate the error to each layer of the text quality network structure to update the network parameters of each layer, and iteratively calculate until the error reaches a preset range;
文本图像质量评估模型生成子模块866,用于获取最终生成的文本图像质量评估模型。The text image quality assessment model generation sub-module 866 is configured to obtain a final generated text image quality assessment model.
本发明实施例提供了一种文本图像质量评估设备,该设备通过获取所需评估的文本图像,将该文本图像输入预先训练的文本图像质量评估模型中,根据文本图像质量评估模型的输出值,确定文本图像的质量指标值,从而通过预先训练的文本图像质量评估模型,能够专门用于对文本图像的质量进行评估,且该评估过程简便易于操作,可以作为OCR前的预处理操作,能够减少计算消耗,与现有技术中的图像质量评估方法相比,大大降低了计算复杂度和计算量,并且与现有技术相比无需对图像进行预处理等操作,评估过程快捷;另外,因为该文本图像质量评估模型是基于深度学习的神经网络进行训练生成,在评估过程中能够模拟人类视觉对文本图像质量的评估过程,且该模型的参数通过迭代进行反复训练,所以通过该预先训练的文本图像质量评估模型对文本图像的质量进行评估,能够提供更加有效和更精准的质量评估结果,提高了评估效率。An embodiment of the present invention provides a text image quality evaluation apparatus, which obtains a text image to be evaluated, inputs the text image into a pre-trained text image quality evaluation model, and estimates an output value of the model according to the text image quality. Determining the quality index value of the text image, so that the pre-trained text image quality evaluation model can be specifically used to evaluate the quality of the text image, and the evaluation process is simple and easy to operate, and can be used as a pre-processing operation before OCR, which can be reduced The calculation consumption is greatly reduced compared with the image quality evaluation method in the prior art, and the calculation complexity and the calculation amount are greatly reduced, and the operation of the image is not required to be pre-processed compared with the prior art, and the evaluation process is quick; The text image quality assessment model is based on the deep learning neural network for training generation. In the evaluation process, the human visual evaluation process of the text image quality can be simulated, and the parameters of the model are iteratively trained repeatedly, so the pre-trained text is passed. Image quality assessment model Image quality assessment, to provide more efficient and more accurate quality assessment, assessment of improved efficiency.
上述所有可选技术方案,可以采用任意结合形成本发明的可选实施例,在此不再一一赘述。All of the above optional technical solutions may be used in any combination to form an optional embodiment of the present invention, and will not be further described herein.
需要说明的是:上述实施例提供的文本图像质量评估设备在执行文本图像质量评估方法时,仅以上述各功能模块的划分进行举例说明,实际应用中,可以根据需要而将上述功能分配由不同的功能模块完成,即将设备的内部结构划分成不同的功能模块,以完成以上描述的全部或者部分功能。另外,上述实施例提供的文本图像质量评估设备与文本图像质量评估方法实施例属于同一构思,其具体实现过程详见方法实施例,这里不再赘述。It is to be noted that the text image quality evaluation device provided in the above embodiment is only illustrated by the division of the above functional modules when performing the text image quality evaluation method. In actual applications, the function distribution may be different according to needs. The function module is completed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the text image quality evaluation device provided by the above embodiment is the same as the text image quality evaluation method embodiment, and the specific implementation process is described in detail in the method embodiment, and details are not described herein again.
本领域普通技术人员可以理解实现上述实施例的全部或部分步骤可以通过硬件来完成,也可以通过程序来指令相关的硬件完成,所述的程序可以存储于一种计算机可读存储介质中,上述提到的存储介质可以是只读存储器,磁盘或光盘等。A person skilled in the art may understand that all or part of the steps of implementing the above embodiments may be completed by hardware, or may be instructed by a program to execute related hardware, and the program may be stored in a computer readable storage medium. The storage medium mentioned may be a read only memory, a magnetic disk or an optical disk or the like.
图16所示为本发明一实施例提供的电子设备的结构示意图。图16提供的电子设备用于执行上述实施例中所提及的图像质量评估方法。如图16所示,该电子设备包括处理器161、存储器162和总线163。FIG. 16 is a schematic structural diagram of an electronic device according to an embodiment of the present invention. The electronic device provided in Fig. 16 is for performing the image quality evaluation method mentioned in the above embodiment. As shown in FIG. 16, the electronic device includes a processor 161, a memory 162, and a bus 163.
处理器161,用于通过总线163调用存储器162中存储的代码,以利用图像样本生成图像质量评估模型,利用图像质量评估模型对待评估图像进行评估操作。The processor 161 is configured to call the code stored in the memory 162 through the bus 163 to generate an image quality evaluation model using the image samples, and perform an evaluation operation on the image to be evaluated by using the image quality evaluation model.
应当理解,该电子设备包括但不限于为手机、平板电脑等电子设备。It should be understood that the electronic device includes, but is not limited to, an electronic device such as a mobile phone or a tablet computer.
在本发明一实施例中,还提供一种计算机存储介质,该计算机可读存储介质上存储有图像质量评估程序,该图像质量评估程序被处理器执行时实现上述任一实施例所提及的图像质量评估方法的操作。In an embodiment of the present invention, a computer storage medium is further provided, where the image quality evaluation program is stored, and when the image quality evaluation program is executed by the processor, the implementation mentioned in any of the above embodiments is implemented. The operation of the image quality assessment method.
应当理解,该计算机可读介质如CD-ROM、软盘、硬盘、数字通用光盘(DVD)、蓝光光盘或其它形式的存储器。替代的,上述实施例提及的图像质量评估方法中的一些操作或所有操作可利用专用集成电路(ASIC)、可编程逻辑器件(PLD)、现场可编程逻辑器件(EPLD)、离散逻辑、硬件、固件等的任意组合被实现。另外,虽然上述实施例的流程图描述了该图像质量评估方法,但可对该图像质量评估方法中的操作进行修改、删除或合并。It should be understood that the computer readable medium is a CD-ROM, a floppy disk, a hard disk, a digital versatile disk (DVD), a Blu-ray disk or other form of memory. Alternatively, some or all of the image quality assessment methods mentioned in the above embodiments may utilize an application specific integrated circuit (ASIC), a programmable logic device (PLD), an on-site programmable logic device (EPLD), discrete logic, hardware. Any combination of firmware, firmware, etc. is implemented. In addition, although the flowchart of the above embodiment describes the image quality evaluation method, the operations in the image quality evaluation method may be modified, deleted, or merged.
如上所述,可利用编码指令(如计算机可读指令)来实现上述任一实施例提及的图像质量评估方法,该编程指令存储于有形计算机可读介质上,如硬盘、闪存、只读存储器(ROM)、光盘(CD)、数字通用光盘(DVD)、高速缓存器、随机访问存储器(RAM)和/或任何其他存储介质,在该存储介质上信息可以存储任意时间(例如,长时间,永久地,短暂的情况,临时缓冲,和/或信息的缓存)。如在此所用的,该术语有形计算机可读介质被明确定义为包括任意类型的计算机可读存储的信号。附加地或替代地,可利用编码指令(如计算机可读指令)实现上述图像质量评估方法实施例所提及的示例过程,该编码指令存储于非暂时性计算机可读介质,如硬盘,闪存,只读存储器,光盘,数字通用光盘,高速缓存器,随机访问存储器和/或任何其他存储介质,在该存储介质信息可以存储任意时间(例如,长时间,永久地,短暂的情况,临时缓冲,和/或信息的缓存)。As described above, the image quality assessment method of any of the above embodiments may be implemented using encoded instructions (such as computer readable instructions) stored on a tangible computer readable medium, such as a hard disk, a flash memory, a read only memory. (ROM), compact disc (CD), digital versatile disc (DVD), cache, random access memory (RAM), and/or any other storage medium on which information can be stored for any time (eg, for a long time, Permanently, short-lived, temporary buffering, and/or caching of information). As used herein, the term tangible computer readable medium is expressly defined to include any type of computer readable stored signal. Additionally or alternatively, the example processes mentioned in the above-described image quality assessment method embodiments may be implemented using encoding instructions (such as computer readable instructions) stored on a non-transitory computer readable medium, such as a hard disk, a flash memory, Read only memory, optical disc, digital versatile disc, cache, random access memory and/or any other storage medium in which information can be stored at any time (eg, long time, permanently, transient, temporary buffering, And/or cache of information).
以上所述仅为本发明的较佳实施例,并不用以限制本发明,凡在本发明的精神和原则之内,所作的任何修改、等同替换、改进等,均应包含在本发明的保护范围之内。The above are only the preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalents, improvements, etc., which are within the spirit and scope of the present invention, should be included in the protection of the present invention. Within the scope.