WO2018082388A1 - 一种肤色检测方法、装置及终端 - Google Patents
一种肤色检测方法、装置及终端 Download PDFInfo
- Publication number
- WO2018082388A1 WO2018082388A1 PCT/CN2017/099869 CN2017099869W WO2018082388A1 WO 2018082388 A1 WO2018082388 A1 WO 2018082388A1 CN 2017099869 W CN2017099869 W CN 2017099869W WO 2018082388 A1 WO2018082388 A1 WO 2018082388A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- skin color
- current frame
- lookup table
- frame picture
- color lookup
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/162—Detection; Localisation; Normalisation using pixel segmentation or colour matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
Definitions
- the present invention relates to the field of image processing technologies, and in particular, to a skin color detecting method, device, and terminal.
- more and more smart phones, tablets and other terminals have video beauty functions, which enable users to add skin, skin whitening and other beauty effects to the faces of the video during video calls and video recording.
- a video with better visual effects Adding a beauty effect to a face first needs to recognize the face area in the picture, and then recognize the skin color area in the face area. Generally, only the skin color area is added to the skin color area, and the non-skin color area (such as eyes and eyebrows is reserved). The authenticity of etc. is not beautified.
- the existing video skin color detection scheme mainly includes the scheme 1: detecting the face information of the current frame picture of the video, and obtaining the approximate contour of the face region according to the Active Shape Model (ASM) algorithm, and estimating the face according to the contour area.
- the skin area avoiding some potentially misleading areas (such as eyes, eyebrows and lip areas), based on the estimated skin color area of the face, based on the pre-set skin threshold empirical parameters, the estimated skin color of the face
- the region performs threshold segmentation to uniformly select a certain number of skin color seeds in different skin color regions. Based on the selected seed points, the spread and detection of the surrounding connected areas are performed, so that all connected skin color regions can be detected.
- Scheme 2 obtaining a face region from a grayscale image of a current frame image of the video, calculating a histogram of the face region, and finding an approximate valley point of the histogram, by using the approximate valley point to the skin color region in the face region and The non-skin area is divided.
- the skin color detection result is discrete, and a visual jump occurs.
- the embodiment of the invention provides a skin color detecting method, device and terminal, which can effectively ensure the continuity of skin color detection results when performing skin color detection on a video.
- a first aspect of the embodiments of the present invention provides a skin color detecting method, including:
- the terminal acquires a video to be detected by the skin color, obtains a current frame image of the video, performs face recognition on the current frame image by using an ASM algorithm, determines a skin color lookup table of the current frame image according to the recognition result, and searches according to the skin color of the current frame image.
- the skin color lookup table accumulated by the historical frame picture before the picture and the current frame picture can obtain the target skin color lookup table with continuous skin color values, and use the target skin color lookup table to detect the skin color of the current frame picture, so that when the skin color is detected for the video, Effectively guarantees the continuity of skin color detection results.
- the terminal may perform weighted average on the skin color lookup table of the current frame picture and the skin color lookup table accumulated by the historical frame picture before the current frame picture to obtain a target skin color lookup table with continuous skin color values, and implement the current frame.
- Automatic learning of skin color values in the skin color lookup table of the picture, and skin color values in the skin color lookup table accumulated for the history frame picture Automatic forgetting, so that when the skin color detection is performed on the video using the target skin color lookup table, continuous skin color detection results can be obtained.
- the manner in which the terminal determines the skin color lookup table of the current frame picture according to the recognition result may be: if the terminal recognizes the face, the terminal determines the face area of the current frame picture according to the first skin color lookup table of the first template picture set.
- the skin color lookup table further determines the skin color lookup table of the current frame picture according to the skin color lookup table of the face area and the second skin color lookup table of the second template picture set.
- the first template picture set includes a number of template pictures that is larger than the number of template pictures included in the second template picture set, and the first skin color lookup table with a wider coverage is used to first determine the skin color lookup table of the face area, that is, the determined person.
- the skin color area in the face area can effectively detect the non-skin color part of the face area.
- the terminal If the terminal does not recognize the face, the terminal directly determines the skin color lookup table of the current frame picture according to the second skin color lookup table, so that the skin color detecting capability is still provided without recognizing the face.
- the terminal may perform filtering and filtering on the current frame image that is detected by the skin color, obtain a mask image, and beautify the current frame image to obtain a beautified current frame image, and then use the mask image to view the current frame image and The landscaping of the current frame image is fused, so that the visual hopping phenomenon can be eliminated when the video is processed according to the continuous skin color detection result, and a good video beauty effect is provided.
- a second aspect of the embodiments of the present invention provides a skin color detecting device, including:
- the obtaining module is configured to acquire a current frame picture of the video for performing skin color detection.
- a determining module configured to perform face recognition on the current frame picture to determine a skin color lookup table of the current frame picture.
- the determining module identifies the human face, determining, according to the first skin color lookup table of the first template picture set, a skin color lookup table of the face area of the current frame picture, and then according to the skin color lookup table of the face area and the second A second skin color lookup table of the template picture set, determining a skin color lookup table of the current frame picture.
- the number of template pictures included in the first template picture set is greater than the number of template pictures included in the second template picture set.
- the skin color lookup table of the current frame picture is determined according to the second skin color lookup table.
- the determining module is further configured to determine the target skin color lookup table according to the skin color lookup table of the current frame picture and the skin color lookup table accumulated by the historical frame picture before the current frame picture.
- the determining module performs weighted averaging on the skin color lookup table of the current frame picture and the skin color lookup table accumulated by the historical frame picture before the current frame picture to obtain a target skin color lookup table.
- the detection module is further configured to perform skin color detection on the current frame picture by using the target skin color lookup table, so that the continuity of the skin color detection result can be effectively ensured when the skin color is detected.
- the device further includes:
- the filtering module is configured to perform a guide filtering on the current frame image that is detected by the skin color to obtain a mask image.
- the beautification module is used to beautify the current frame picture to obtain a picture frame of the current frame that has been beautified.
- the video can be visually processed to eliminate the visual jump phenomenon and provide a good video beauty effect.
- a third aspect of the embodiments of the present invention provides a terminal, including: a processor and a memory, where the processor and the memory are connected by a bus, the memory stores executable program code, and the processor is configured to call executable program code in the memory to execute The skin color detecting method described in any one of the above first aspects.
- the current frame picture of the video for detecting the skin color is obtained, and the current frame picture is subjected to face recognition to determine a skin color lookup table of the current frame picture, and according to the skin color lookup table of the current frame picture and the current a skin color lookup table accumulated by the history frame picture before the frame picture, determining a target skin color lookup table with continuous skin color values, thereby performing skin color detection on the current frame picture by using the target skin color lookup table, thereby performing skin color detection on the video Effectively guarantees the continuity of skin color detection results.
- FIG. 1 is a schematic structural diagram of a terminal according to an embodiment of the present invention.
- FIG. 2 is a schematic flow chart of a first embodiment of a skin color detecting method according to an embodiment of the present invention
- FIG. 3 is a schematic flow chart of a second embodiment of a skin color detecting method according to an embodiment of the present invention.
- FIG. 4 is a schematic diagram of a skin color calibration method according to an embodiment of the present invention.
- FIG. 5 is a schematic structural diagram of a skin color detecting device according to an embodiment of the present invention.
- the terminal described in the embodiment of the present invention may specifically include, but is not limited to, a smart phone, a tablet computer, a digital camera, a mobile Internet device (MID), and the like.
- a smart phone a tablet computer
- a digital camera a digital camera
- a mobile Internet device MID
- FIG. 1 is a schematic structural diagram of a terminal according to an embodiment of the present invention.
- the terminal described in this embodiment includes a processor 101, a memory 102, an output device 103, and an input device 104.
- the processor 101 is connected to the memory 102, the output device 103, and the input device 104 via a bus.
- the processor 101 may be a baseband processor, a baseband chip, a digital signal processor (DSP), or a system on chip (SOC) including a baseband processor and an application processor.
- the above memory 102 is a memory device of the terminal for storing programs and data. It can be understood that the memory 102 herein may be a high-speed RAM memory, or may be a non-volatile memory, such as at least one disk memory; optionally, at least one of the processors 101 may be located away from the foregoing processor 101.
- the output device 103 described above can be a display.
- the input device 104 may be a touch panel, a camera, a microphone, or the like.
- the memory 102 is configured to store a set of program codes, and the processor 101 calls the program code stored in the memory 102 to perform the following operations:
- the processor 101 acquires a video to be subjected to skin color detection, and acquires a current frame picture of the video.
- the processor 101 performs face recognition on the current frame picture to obtain a face recognition result, and determines a skin color lookup table of the current frame picture according to the face recognition result.
- the skin color lookup table that is, the 3D lookup table, stores the correspondence between the pixel values of the pixels and the skin color values, and the pixel values are red, green, blue, and RGB values, and the structure of the skin color lookup table may be a subscript. For the pixel value, the content corresponding to the subscript is the skin color value.
- the skin color lookup table can determine whether a pixel is a skin color. For any pixel, the pixel value is obtained, the skin color value corresponding to the pixel value is queried from the skin color lookup table, and the skin color value corresponding to the skin color and the non skin color according to the predetermined skin color value is used. The situation determines if the pixel is skin tone.
- the processor 101 performs face recognition on the current frame picture, and if the face is recognized, determines a skin color lookup table of the face area of the current frame picture according to the first skin color lookup table of the first template picture set, and then according to the a skin color lookup table of the face area and a second skin color lookup table of the second template picture set, determining a skin color lookup table of the current frame picture; if no face is recognized, the skin color of the current frame picture may be directly determined by the second skin color lookup table Lookup table.
- the number of template pictures included in the first template picture set is greater than the number of template pictures included in the second template picture set.
- the processor 101 determines a target skin color lookup table for performing skin color detection on the current frame picture according to the skin color lookup table of the current frame picture and the skin color lookup table accumulated by the historical frame picture before the current frame picture.
- the processor 101 performs skin color detection on the current frame picture using the target skin color lookup table.
- the processor 101 obtains a pixel value for each pixel in the current frame picture, and queries the skin color value corresponding to the pixel value from the target skin color lookup table, and then according to the preset skin color value and the skin color and the non-skin color. Corresponding to the situation can determine whether each pixel is skin color. Alternatively, the processor 101 may determine, from the target skin color lookup table, a target pixel value corresponding to the skin color corresponding to the skin color value, determine a pixel whose pixel value is the target pixel value as the skin color, and determine other pixels as the non-skin color.
- the processor 101 performs direction filtering on the current frame image that is detected by the skin color to obtain a mask image, and performs beautification processing on the current frame image to obtain a beautified current frame image, and then uses the mask image to view the current frame image and The merging of the current frame picture is fused.
- the mask image is used as the mask image.
- the role of the mask is to convert different grayscale values into different transparency, and apply to the layer where it is located, so that the transparency of different parts of the layer changes accordingly.
- white is completely opaque and gray is translucent.
- Fusion is a blend of transparency to get the overlay of two images.
- the processor 101, the memory device 102, the output device 103, and the input device 104 described in the embodiments of the present invention may perform the first embodiment and the second embodiment of the skin color detecting method provided by the embodiment of the present invention.
- the implementation of the described terminal can also implement the implementation of the skin color detecting device described in the skin color detecting device provided by the embodiment of the present invention, and details are not described herein again.
- the terminal acquires a current frame image of the video for detecting the skin color, and performs a human image on the current frame image. Face recognition, determining a skin color lookup table of the current frame picture, and determining a target skin color lookup table with continuous skin color values according to the skin color lookup table of the current frame picture and the skin color lookup table accumulated by the historical frame picture before the current frame picture Therefore, the skin color detection is performed on the current frame picture by using the target skin color lookup table, so that the continuity of the skin color detection result can be effectively ensured when the skin color is detected.
- FIG. 2 is a schematic flowchart diagram of a first embodiment of a skin color detecting method according to an embodiment of the present invention.
- the skin color detecting method described in this embodiment includes the following steps:
- the terminal acquires a current frame picture of the video that performs skin color detection.
- the terminal acquires a video to be detected by the skin color, and acquires a current frame picture of the video.
- the terminal performs face recognition on the current frame picture to determine a skin color lookup table of the current frame picture.
- the skin color lookup table that is, the 3D lookup table, stores the correspondence between the pixel value of the pixel and the skin color value, and the pixel value is the RGB value.
- the structure of the skin color lookup table may be a subscript as a pixel value, and the content corresponding to the subscript is the skin color. value.
- the skin color lookup table can determine whether a pixel is a skin color. For any pixel, the pixel value is obtained, the skin color value corresponding to the pixel value is queried from the skin color lookup table, and the skin color value corresponding to the skin color and the non skin color according to the predetermined skin color value is used. The case determines whether the pixel is a skin color.
- the skin color value may have a first value and a second value, and the pixel value corresponding to the pixel value is the skin color of the first value, and the skin color value corresponding to the pixel value is the second value.
- the pixels are not skin tones.
- Each skin color lookup table may be initialized to a state in which all pixels are non-skinned, that is, the skin color value is null or the second value.
- the RGB depth is 8 bits, the skin color value is 1 for skin color, and the skin color value is 0 for non-skin color.
- the pixel value ranges from 0 to 16777215 (2 24 -1).
- the skin color lookup table may be specifically as shown in Table 1.
- the pixel value and the skin color value corresponding to the pixel value are stored.
- the pixel value is m. s, when the corresponding skin color value is 1, the pixel whose pixel value is m or s is the skin color, and when the pixel value is 0, n, 2 24 -1, the corresponding skin color value is 0, indicating that the pixel value is 0 or
- the pixels of n or 2 24 -1 are non-skin.
- the specific manner in which the terminal performs face recognition on the current frame picture to determine the skin color lookup table of the current frame picture may include, but is not limited to:
- Manner 1 Detect the face information of the current frame picture, obtain the approximate outline of the face area according to the ASM algorithm, estimate the skin area of the face according to the outline area, and avoid some potentially misleading areas (such as eyes, eyebrows and lip areas).
- the skin color region of the estimated face is subjected to threshold segmentation according to the skin threshold empirical parameter set in advance, and a certain number of skin color seeds are uniformly selected in different skin color regions.
- the selected seed point the spread and detection of the surrounding connected area are performed, so that all connected skin color regions can be detected, and the pixel values of each pixel in all connected skin color regions can be corresponding to the skin color lookup table of the current frame picture.
- the skin color value is set to the first value, thereby determining the skin color lookup table of the current frame picture.
- Manner 2 obtaining a face region from the grayscale image of the current frame picture, calculating a histogram of the face region, and finding an approximate valley point of the histogram, and the skin color region and the non-face region in the face region are obtained by the approximate valley point Skin color area is divided, face will be The pixel value of each pixel in the skin color region in the region is set to a first value in the skin color lookup table of the current frame image, thereby determining the skin color lookup table of the current frame image.
- the terminal determines, according to the skin color lookup table of the current frame picture and the skin color lookup table accumulated by the historical frame picture before the current frame picture, to determine the target skin color lookup table.
- the terminal performs skin color detection on the current frame picture by using the target skin color lookup table.
- the skin color lookup table realizes automatic learning of skin color values in the skin color lookup table of the current frame picture, and automatic forgetting of skin color values in the skin color lookup table accumulated for the history frame picture.
- the first value may take 255
- the second value may take 0.
- the skin color value in the skin color lookup table of the current frame picture is discrete, and is 255 (skin tone) or 0 (non-skin tone), weighted.
- the skin color values in the target skin lookup table are continuous with a range of [0, 255].
- the terminal acquires the pixel value of each pixel in the current frame picture, and queries the skin color value corresponding to the pixel value of each pixel from the target skin color lookup table of the current frame picture, and the pixel corresponding to the pixel value corresponding to the pixel value is determined to be the skin color.
- a pixel whose skin color value is 0 corresponding to a pixel value is determined to be non-skin tone, and a skin color value other than 0 and 255 is a skin color value, and the higher the confidence level, the closer the corresponding pixel is to the skin color, thereby accurately
- the skin color area of the current frame picture is detected. It can be seen that the skin color detection of each frame of the video is performed by using the skin color lookup table with continuous skin color values, so that the detection result of the skin color detection of the video is continuous.
- the terminal acquires a current frame picture of the video for detecting the skin color, performs face recognition on the current frame picture, determines a skin color lookup table of the current frame picture, and performs a skin color lookup table according to the current frame picture and a skin color lookup table accumulated by the history frame picture before the current frame picture, determining a target skin color lookup table with continuous skin color values, thereby performing skin color detection on the current frame picture by using the target skin color lookup table, thereby performing skin color detection on the video , can effectively ensure the continuity of skin color detection results.
- FIG. 3 is a schematic flowchart diagram of a second embodiment of a skin color detecting method according to an embodiment of the present invention.
- the skin color detecting method described in this embodiment includes the following steps:
- the terminal acquires a current frame picture of a video for performing skin color detection.
- the terminal may separately perform skin color calibration on the two template picture sets (ie, the first template picture set and the second template picture set) to obtain two skin color lookup tables (ie, the first skin color lookup table and the first
- the second skin tone lookup table, the first skin tone lookup table and the second skin tone lookup table are used for skin color detection of the video.
- the first skin color lookup table corresponds to the first template picture set
- the second skin color lookup table corresponds to the second template picture set
- the terminal performs skin color calibration on the first template picture set and the second template picture set respectively to obtain the first skin color lookup table.
- the specific steps of the second skin color lookup table can be as follows:
- the skin color area in the template picture is selected by selecting a frame, as shown in the rectangular selection box in FIG. 4, and the rectangle is selected by using different areas in the template picture.
- Frame the box to select the skin color area (such as face, neck, arms, hands, legs, etc.).
- For each picture area selected in the selected box obtain the total number of pixels included in the selected image area of the selected frame.
- the pixel value of each pixel in order to facilitate interpolation of pixels in the current frame picture to obtain a skin color detection result of continuous skin color values, each of the selected image areas included in the selected frame
- the pixel value of one pixel loses the precision of the preset number of bits. In this embodiment, the loss of the 3-bit precision is taken as an example.
- the pixel value is increased by 4, and then the right bit is shifted by 3 bits, and the number of pixels corresponding to each pixel value in the selected picture area is counted. Determining, in all the template images included in the first template picture set, the corresponding pixel number is greater than or equal to a first target pixel value of a preset proportion (for example, 1%) of the total number of pixels, it being understood that the first target pixel value is specific Includes multiple pixel values.
- a preset proportion for example, 1%) of the total number of pixels
- the pixel whose pixel value is the first target pixel value is regarded as the skin color
- the pixel whose pixel value is the other pixel value is regarded as the non-skin color
- the skin color value corresponding to the first target pixel value in the first skin color lookup table is set as the first value.
- the skin color value corresponding to the pixel value other than the first target pixel value in the first skin color lookup table is set to the second value (ie, non-skin tone), thereby completing the determination of the first skin color lookup table.
- the skin color value may be set for the skin color lookup table of each template image included in the first template picture set, and then the skin color lookup table of each template picture is superimposed to obtain
- the first skin color lookup table may be: for each template image included in the first template picture set, determining that the corresponding pixel number is greater than or equal to a preset ratio (for example, 1%) of the total number of pixels, and The skin color value corresponding to the first target pixel value in the skin color lookup table of the template picture is set to a first value, and the skin color value corresponding to the pixel value other than the first target pixel value is set to a second value, and finally each template is
- the skin color lookup table of the picture is superimposed, which may be a method of taking a union, that is, for the same pixel value, as long as the skin color value in the skin color lookup table of one template picture is the first value, the first template picture set is The corresponding skin color value in the first skin color lookup table is set to a first
- the determination of the second skin color lookup table of the second template picture set can be completed in the same manner as described above.
- the difference between the first template picture set and the second template picture set is that the number of template pictures included in the first template picture set is greater than the number of template pictures included in the second template picture set, that is, the coverage of the first template picture set. More broadly, skin color scaling can be performed on more pixel values than the second template image set.
- the first template picture set may specifically include a local picture captured by the terminal under a specific camera parameter, and a picture on the Internet.
- the second template picture set may specifically include only the local picture captured by the terminal under the specific picture parameter.
- the specific photo parameters corresponding to the first template picture set may be multi-color temperature, automatic exposure (AE), automatic white balance (AWB), etc., in order to capture pictures at various color temperatures, without missing detection
- the specific photo parameters corresponding to the second template picture set may be normal color temperature, AE, AWB, etc., in order to capture pictures under normal color temperature, without the principle of false detection.
- the skin color area of the face is larger and representative than the other parts of the human body
- the first template picture set and the second template picture set include
- the template image may be a picture including at least a human face.
- the terminal performs face recognition on the current frame picture. If the face is recognized, determining a skin color lookup table of the face area of the current frame picture according to the first skin color lookup table of the first template picture set.
- determining, according to the first skin color lookup table, the skin color lookup table of the face region including: acquiring pixel values of each pixel in the face region, for each The pixel value of the pixel also loses the 3-bit precision.
- the skin color value corresponding to the pixel value after the pixel precision is lost for each pixel in the face region is queried, if the skin color value corresponding to the first pixel value is the first A value, the pixel whose pixel value is the first pixel value after the loss of the 3-bit precision is the skin color, and the skin color value corresponding to the first pixel value in the skin color lookup table of the face region is also set to the first value, for the loss of 3 bits.
- the corresponding skin color value is the pixel value of the second value
- the corresponding skin color value in the skin color lookup table of the face area is also set to the second value, and the first skin color lookup table with wider coverage is used first.
- the skin color lookup table of the area that is, the skin color area in the face area, can effectively detect non-skinned parts in the face area, such as eyes, lips, glasses, eyebrows, etc., and can also avoid deviation of face recognition ( For example, if the recognized face area is larger than the actual area, the skin color is falsely detected.
- the terminal determines a skin color lookup table of the current frame picture according to the skin color lookup table of the face area and the second skin color lookup table of the second template picture set.
- the skin color lookup table and the second skin color lookup table of the determined face region are determined by taking a union, and the skin color lookup table of the current frame image is determined, including: a skin color lookup table of the face region and the second In the skin color lookup table, if the skin color value corresponding to the same pixel value has a first value or a first value, the same pixel value may be set to the corresponding skin color value in the skin color lookup table of the current frame picture.
- the skin color value corresponding to the same pixel value is the second value in the skin color lookup table and the second skin color lookup table of the face region, the same pixel value corresponds to the skin color lookup table of the current frame image
- the skin color value is set to the second value, and the skin color lookup table of the current frame picture is determined by the first skin color lookup table, the second skin color lookup table, and the face recognition, which can greatly reduce the missed detection rate and the false detection rate.
- the terminal may directly determine the skin color lookup table of the current frame picture according to the second skin color lookup table with a smaller coverage, including: acquiring the current frame.
- the pixel value of each pixel in the picture is used to query the skin color value corresponding to the pixel value of each pixel in the current frame picture from the second skin color lookup table, and if the skin color value corresponding to the second pixel value is the first value, the pixel value
- the pixel of the second pixel value is the skin color
- the skin color value corresponding to the second pixel value in the skin color lookup table of the current frame picture is also set to the first value
- the pixel value of the second value is The corresponding skin color value in the skin color lookup table of the current frame picture is also set to the second value.
- the second skin color lookup table is determined according to the principle that the terminal does not detect by mistake, and the second skin color lookup table has better skin color detection capability when the photographing parameter is normal color temperature, AE, AWB. It is possible to make the skin color detecting ability still available in the case where the face is not recognized from the current frame picture.
- the terminal determines, according to the skin color lookup table of the current frame picture and the skin color lookup table accumulated by the historical frame picture before the current frame picture, to determine the target skin color lookup table.
- the terminal performs skin color detection on the current frame picture by using the target skin color lookup table.
- the terminal acquires the pixel value of each pixel in the current frame image, and the pixel value of each pixel also loses 3 digits of precision. From the target skin color lookup table, the pixel value corresponding to each pixel is lost by 3 digits. The skin color value so that the skin color area of the current frame picture can be accurately detected.
- the terminal may also interpolate each pixel in the current frame picture by using the target skin color lookup table to obtain a skin color detection result with continuous skin color values.
- the tetrahedral linear interpolation is performed on each pixel in the current frame picture.
- the steps of linear interpolation of the tetrahedron are as follows: 8 on the R/G/B three axes respectively.
- the entire color space is divided into 32768 uniform small cubes (the cube side length is 8), and each uniform cube is divided into six tetrahedrons without any overlap except for the surface overlap according to a specific rule.
- the skin color lookup table of the current frame picture the skin color values of the four vertices can be known, and the skin color value of the pixel can be inserted by linear interpolation.
- the terminal performs direction filtering on the current frame image that is detected by the skin color to obtain a mask image.
- the steering filter has two inputs, one is the input graph p, one is the steering graph I, and one has an output q.
- p is an interpolation result
- I is a grayscale image or a single-channel image of the input image (ie, the current frame image)
- q is an optimized skin color detection result map.
- the steering filtering is based on the local linear model.
- the image is considered to be a two-dimensional function, and the analytical expression cannot be written. Therefore, it is assumed that the input (direction map) of the function and the output satisfy the linear relationship in a window as follows:
- the calculation step of the output q can be as follows:
- the function of fmean is the average value of window pixels with radius 20, and the role of e is to make the divisor not 0, e The smaller the value, the better.
- the terminal performs beautification processing on the current frame picture to obtain a picture frame of the current frame that is beautified.
- the terminal uses the mask image to merge the current frame image with the beautified current frame image.
- ⁇ is the mask image
- image 1 is the current frame image
- image 2 is the landscaping current frame image
- result is the final processing result
- the terminal acquires a current frame image of the video for detecting the skin color, and determines a skin color lookup table of the face region of the current frame image according to the first skin color lookup table of the first template image set, according to the skin color of the face region.
- Finding a second skin color lookup table of the table and the second template picture set determining a skin color lookup table of the current frame picture, determining a skin color according to the skin color lookup table of the current frame picture and the skin color lookup table accumulated by the historical frame picture before the current frame picture
- the target skin color lookup table with continuous values is used to detect the skin color of the current frame image by using the target skin color lookup table, and the current frame image detected by the skin color is guided and filtered to obtain a mask image, and the current frame image is beautified to obtain a beautified
- the current frame image is merged with the current frame image and the beautified current frame image by using the mask image, so that when the skin color is detected, the continuity of the skin color detection result can be effectively ensured, and then the video is detected according to the continuous skin color detection result. It can eliminate visual jumps when making beauty treatments, providing good The beauty of video effects.
- FIG. 5 is a schematic structural diagram of a skin color detecting device according to an embodiment of the present invention.
- the skin color detecting device described in this embodiment includes:
- the obtaining module 501 is configured to acquire a current frame picture of the video for performing skin color detection.
- the determining module 502 is configured to perform face recognition on the current frame picture to determine a skin color lookup table of the current frame picture.
- the determining module 502 identifies the human face, determining, according to the first skin color lookup table of the first template picture set, a skin color lookup table of the face area of the current frame picture, and then according to the skin color lookup table of the face area and the first A second skin color lookup table of the second template picture set, determining a skin color lookup table of the current frame picture.
- the number of template pictures included in the first template picture set is greater than the number of template pictures included in the second template picture set.
- the determining module 502 determines the skin color lookup table of the current frame picture according to the second skin color lookup table if the human face is not recognized.
- the determining module 502 is further configured to determine the target skin color lookup table according to the skin color lookup table of the current frame picture and the skin color lookup table accumulated by the historical frame picture before the current frame picture.
- the determining module 502 performs weighted averaging on the skin color lookup table of the current frame picture and the skin color lookup table accumulated by the historical frame picture before the current frame picture to obtain a target skin color lookup table.
- the detecting module 503 is further configured to perform skin color detection on the current frame picture by using the target skin color lookup table.
- the device further includes:
- the filtering module 504 is configured to perform a guide filtering on the current frame image that is detected by the skin color to obtain a mask image.
- the beautification module 505 is configured to beautify the current frame picture to obtain a beautified current frame picture.
- the fusion module 506 is configured to fuse the current frame image and the beautified current frame image by using the mask image.
- the terminal acquires a current frame picture of the video for detecting the skin color, performs face recognition on the current frame picture, determines a skin color lookup table of the current frame picture, and performs a skin color lookup table according to the current frame picture and a skin color lookup table accumulated by the history frame picture before the current frame picture, determining a target skin color lookup table with continuous skin color values, thereby performing skin color detection on the current frame picture by using the target skin color lookup table, thereby performing skin color detection on the video , can effectively ensure the continuity of skin color detection results.
- the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
本发明实施例提供了一种肤色检测方法、装置及终端，其中方法包括：获取进行肤色检测的视频的当前帧图片；对该当前帧图片进行人脸识别，以确定该当前帧图片的肤色查找表；根据该当前帧图片的肤色查找表和该当前帧图片之前的历史帧图片累计的肤色查找表，确定出目标肤色查找表；利用该目标肤色查找表对该当前帧图片进行肤色检测。通过本发明实施例在对视频进行肤色检测时，可以有效保证肤色检测结果的连续性。
Description
本申请要求于2016年11月2日提交中国专利局、申请号为201610950691.0、发明名称为“一种肤色检测方法、装置及终端”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
本发明涉及图像处理技术领域，具体涉及一种肤色检测方法、装置及终端。
目前，越来越多的智能手机、平板电脑等终端设有视频美颜功能，使得用户在视频通话、拍摄视频时，可以对视频中的人脸加入磨皮、美白等美颜效果，以得到人物视觉效果较好的视频。对人脸加入美颜效果首先需要识别出图片中的人脸区域，再识别出人脸区域中的肤色区域，一般只需对肤色区域加入美颜效果，而保留非肤色区域(例如眼睛、眉毛等)的真实性而不予美化。
现有的视频肤色检测方案主要有，方案一：检测视频当前帧图片的人脸信息，根据主动形状模型(Active Shape Model，ASM)算法，得到人脸区域的大概轮廓，依据轮廓区域估计人脸的皮肤区域，规避掉一些可能误导的区域(比如眼睛，眉毛和嘴唇区域)，依据估计出的人脸的肤色区域，根据事先设定好的皮肤阈值经验参数，对估计出的人脸的肤色区域进行阈值分割，在不同的肤色区域，均匀选择一定数量的肤色种子。根据选定的种子点，进行周边连通区域的蔓延和检测，从而可以检测到所有连通的肤色区域。方案二：从视频当前帧图片的灰度图中获取人脸区域，计算人脸区域的直方图，并找到该直方图的近似谷点，通过该近似谷点对人脸区域中的肤色区域和非肤色区域进行划分。然而，上述两种方案在视频中识别出人脸和未识别出人脸时肤色检测结果是离散的，会产生视觉上的跳变。
发明内容
本发明实施例提供了一种肤色检测方法、装置及终端，在对视频进行肤色检测时，可以有效保证肤色检测结果的连续性。
本发明实施例第一方面提供了一种肤色检测方法，包括：
终端获取待进行肤色检测的视频，并获取该视频的当前帧图片，利用ASM算法等对当前帧图片进行人脸识别，根据识别结果确定当前帧图片的肤色查找表，根据当前帧图片的肤色查找表和当前帧图片之前的历史帧图片累计的肤色查找表可以得到肤色值连续的目标肤色查找表，并利用目标肤色查找表对当前帧图片进行肤色检测，从而在对视频进行肤色检测时，可以有效保证肤色检测结果的连续性。
可选的，终端可以通过对当前帧图片的肤色查找表和当前帧图片之前的历史帧图片累计的肤色查找表进行加权平均的方式，得到肤色值连续的目标肤色查找表，实现了对当前帧图片的肤色查找表中的肤色值的自动学习，以及对历史帧图片累计的肤色查找表中的肤色值的
自动遗忘，使得利用该目标肤色查找表对视频进行肤色检测时，可以得到连续的肤色检测结果。
其中，目标肤色查找表中的肤色值＝(1-ω)*历史帧图片累计的肤色查找表中的肤色值+ω*当前帧图片的肤色查找表中的肤色值，ω为加权系数，一般可取ω＝5％。
可选的，终端根据识别结果确定当前帧图片的肤色查找表的方式可以为：终端若识别到人脸，则根据第一模板图片集的第一肤色查找表，确定当前帧图片的人脸区域的肤色查找表，再根据人脸区域的肤色查找表和第二模板图片集的第二肤色查找表，确定当前帧图片的肤色查找表。其中，第一模板图片集包括的模板图片的数量大于第二模板图片集包括的模板图片的数量，利用覆盖范围更广的第一肤色查找表先确定人脸区域的肤色查找表，即确定人脸区域中的肤色区域，可以有效检测出人脸区域中的非肤色部分。
终端若未识别到人脸，则直接根据第二肤色查找表，确定当前帧图片的肤色查找表，使得在没有识别到人脸的情况下仍然具备肤色检测能力。
可选的，终端可以对经过肤色检测的当前帧图片进行导向滤波，得到蒙版图片，并对当前帧图片进行美化处理，得到经过美化的当前帧图片，再利用蒙版图片将当前帧图片和经过美化的当前帧图片进行融合，从而根据连续的肤色检测结果对视频进行美颜处理时可以消除视觉上的跳变现象，提供良好的视频美颜效果。
本发明实施例第二方面提供了一种肤色检测装置，包括：
获取模块，用于获取进行肤色检测的视频的当前帧图片。
确定模块，用于对当前帧图片进行人脸识别，以确定当前帧图片的肤色查找表。
具体的，确定模块若识别到人脸，则根据第一模板图片集的第一肤色查找表，确定当前帧图片的人脸区域的肤色查找表，再根据人脸区域的肤色查找表和第二模板图片集的第二肤色查找表，确定当前帧图片的肤色查找表。
其中，第一模板图片集包括的模板图片的数量大于第二模板图片集包括的模板图片的数量。
确定模块若未识别到人脸，则根据第二肤色查找表，确定当前帧图片的肤色查找表。
确定模块，还用于根据当前帧图片的肤色查找表和当前帧图片之前的历史帧图片累计的肤色查找表，确定出目标肤色查找表。
具体的，确定模块对当前帧图片的肤色查找表和所述当前帧图片之前的历史帧图片累计的肤色查找表进行加权平均，得到目标肤色查找表。
其中，目标肤色查找表中的肤色值＝(1-ω)*历史帧图片累计的肤色查找表中的肤色值+ω*当前帧图片的肤色查找表中的肤色值，ω为加权系数，一般可取ω＝5％。
检测模块，还用于利用目标肤色查找表对当前帧图片进行肤色检测，从而在对视频进行肤色检测时，可以有效保证肤色检测结果的连续性。
可选的，该装置还包括：
滤波模块，用于对经过肤色检测的当前帧图片进行导向滤波，得到蒙版图片。
美化模块，用于对当前帧图片进行美化处理，得到经过美化的当前帧图片。
融合模块，用于利用蒙版图片将当前帧图片和经过美化的当前帧图片进行融合，从而根
据连续的肤色检测结果对视频进行美颜处理时可以消除视觉上的跳变现象，提供良好的视频美颜效果。
本发明实施例第三方面提供了一种终端，包括：处理器和储存器，处理器和存储器通过总线连接，存储器存储有可执行程序代码，处理器用于调用存储器中的可执行程序代码，执行上述第一方面中任一项所描述的肤色检测方法。
本发明实施例通过获取进行肤色检测的视频的当前帧图片，对该当前帧图片进行人脸识别，以确定该当前帧图片的肤色查找表，并根据该当前帧图片的肤色查找表和该当前帧图片之前的历史帧图片累计的肤色查找表，确定出肤色值连续的目标肤色查找表，从而利用该目标肤色查找表对该当前帧图片进行肤色检测，从而在对视频进行肤色检测时，可以有效保证肤色检测结果的连续性。
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。
图1是本发明实施例提供的一种终端的结构示意图；
图2是本发明实施例提供的一种肤色检测方法的第一实施例流程示意图；
图3是本发明实施例提供的一种肤色检测方法的第二实施例流程示意图；
图4是本发明实施例提供的一种肤色标定方式的示意图；
图5是本发明实施例提供的一种肤色检测装置的结构示意图。
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作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。
本发明实施例中所描述的终端具体可以包括但不限于：智能手机、平板电脑、数码相机、移动互联网设备(Mobile Internet Device，MID)等。
请参阅图1，为本发明实施例提供的一种终端的结构示意图。本实施例中所描述的终端，包括：处理器101、存储器102、输出设备103和输入设备104，上述处理器101通过总线与存储器102、输出设备103和输入设备104连接。
其中，上述处理器101具体可以为基带处理器、基带芯片、数字信号处理器(Digital Signal Processor，DSP)或者包括基带处理器和应用处理器在内的片上***(System on Chip，SOC)等。上述存储器102是终端的记忆设备，用于存放程序和数据。可以理解的是，此处的存储器102可以是高速RAM存储器，也可以是非不稳定的存储器(non-volatile memory)，例如至少一个磁盘存储器；可选的还可以是至少一个位于远离前述处理器101的存储装置。上述输出设备103可以为显示器。上述输入设备104可以为触控面板、摄像头、麦克风等。
上述存储器102，用于存储一组程序代码，上述处理器101调用存储器102中存储的程序代码，执行如下操作：
处理器101获取待进行肤色检测的视频，并获取该视频的当前帧图片。
处理器101对当前帧图片进行人脸识别，得到人脸识别结果，并根据人脸识别结果确定当前帧图片的肤色查找表。
其中，肤色查找表也即3D查找表，存储有像素的像素值与肤色值的对应关系，像素值即红绿蓝(Red、Green、Blue，RGB)值，肤色查找表的构造可以是下标为像素值，下标对应的内容为肤色值。通过肤色查找表可以确定一个像素是否为肤色，对于任意一个像素，获取其像素值，从肤色查找表中查询其像素值对应的肤色值，再根据事先规定的肤色值与肤色、非肤色的对应情况确定该像素是否为肤色。
具体的，处理器101对当前帧图片进行人脸识别，如果识别到人脸，则根据第一模板图片集的第一肤色查找表，确定当前帧图片的人脸区域的肤色查找表，再根据人脸区域的肤色查找表和第二模板图片集的第二肤色查找表，确定当前帧图片的肤色查找表；如果没有识别到人脸，则可以直接第二肤色查找表确定当前帧图片的肤色查找表。
其中，第一模板图片集包括的模板图片的数量大于第二模板图片集包括的模板图片的数量。
处理器101根据当前帧图片的肤色查找表和当前帧图片之前的历史帧图片累计的肤色查找表，确定用于对当前帧图片进行肤色检测的目标肤色查找表。
具体的，处理器101可以对当前帧图片的肤色查找表和当前帧图片之前的历史帧图片累计的肤色查找表进行加权平均，从而得到用于对当前帧图片进行肤色检测的目标肤色查找表。即目标肤色查找表中的肤色值＝(1-ω)*历史帧图片累计的肤色查找表中的肤色值+ω*当前帧图片的肤色查找表中的肤色值，ω为加权系数。
处理器101利用目标肤色查找表对当前帧图片进行肤色检测。
具体的，处理器101针对当前帧图片中的每一个像素，获取其像素值，从目标肤色查找表中，查询该像素值对应的肤色值，再根据事先规定的肤色值与肤色、非肤色的对应情况即可确定每一个像素是否为肤色。或者，处理器101也可以从目标肤色查找表中，确定出对应的肤色值对应肤色的目标像素值，将像素值为该目标像素值的像素确定为肤色，将其它像素确定为非肤色。
进一步的，处理器101对经过肤色检测的当前帧图片进行导向滤波，得到蒙版图片，对当前帧图片进行美化处理，得到经过美化的当前帧图片，再利用该蒙版图片将当前帧图片和经过美化的当前帧图片进行融合。
其中，蒙版图片即作为蒙版的图片，蒙版的作用是将不同灰度色值转化为不同的透明度，并作用到它所在的图层，使图层不同部位透明度产生相应的变化，黑色为完全透明，白色为完全不透明，灰色为半透明。融合即进行透明度混合，以获得两张图片的叠加效果。
具体实现中，本发明实施例中所描述的处理器101、存储器102、输出设备103和输入设备104可执行本发明实施例提供的一种肤色检测方法的第一实施例和第二实施例中所描述的终端的实现方式，也可执行本发明实施例提供的一种肤色检测装置中所描述的肤色检测装置的实现方式，在此不再赘述。
本发明实施例中，终端获取进行肤色检测的视频的当前帧图片，对该当前帧图片进行人
脸识别，以确定该当前帧图片的肤色查找表，并根据该当前帧图片的肤色查找表和该当前帧图片之前的历史帧图片累计的肤色查找表，确定出肤色值连续的目标肤色查找表，从而利用该目标肤色查找表对该当前帧图片进行肤色检测，从而在对视频进行肤色检测时，可以有效保证肤色检测结果的连续性。
请参阅图2，为本发明实施例提供的一种肤色检测方法的第一实施例流程示意图。本实施例中所描述的肤色检测方法，包括以下步骤：
201、终端获取进行肤色检测的视频的当前帧图片。
具体的，终端获取待进行肤色检测的视频，并获取该视频的当前帧图片。
202、终端对当前帧图片进行人脸识别，以确定当前帧图片的肤色查找表。
其中，肤色查找表也即3D查找表，存储有像素的像素值与肤色值的对应关系，像素值即RGB值，肤色查找表的构造可以是下标为像素值，下标对应的内容为肤色值。通过肤色查找表可以确定一个像素是否为肤色，对于任意一个像素，获取其像素值，从肤色查找表中查询其像素值对应的肤色值，再根据事先规定的肤色值与肤色、非肤色的对应情况确定该像素是否为肤色，例如，可以设定肤色值有第一数值、第二数值，像素值对应的肤色值为第一数值的像素为肤色，像素值对应的肤色值为第二数值的像素为非肤色。其中，各个肤色查找表可以初始化为所有像素均为非肤色的状态，即肤色值为空或者第二数值。以RGB深度是8比特，肤色值为1表示肤色，肤色值为0表示非肤色为例，则像素值的数量为28*28*28＝224＝16777216个，以十进制形式表示时，像素值的范围是：0～16777215(224-1)，肤色查找表具体可以如表1所示，存储有像素值以及像素值对应的肤色值，在表1中，像素值为m、s时，对应的肤色值为1，则表示像素值为m或s的像素为肤色，像素值为0、n、224-1时，对应的肤色值为0，则表示像素值为0或n或224-1的像素为非肤色。
| 像素值 | 0 | …… | m | …… | n | …… | s | …… | 224-1 |
| 肤色值 | 0 | …… | 1 | …… | 0 | …… | 1 | …… | 0 |
表1
其中，终端通过对当前帧图片进行人脸识别以确定当前帧图片的肤色查找表的具体方式可以包括但不限于：
方式一：检测当前帧图片的人脸信息，根据ASM算法，得到人脸区域的大概轮廓，依据轮廓区域估计人脸的皮肤区域，规避掉一些可能误导的区域(比如眼睛，眉毛和嘴唇区域)，依据估计出的人脸的肤色区域，根据事先设定好的皮肤阈值经验参数，对估计出的人脸的肤色区域进行阈值分割，在不同的肤色区域，均匀选择一定数量的肤色种子。根据选定的种子点，进行周边连通区域的蔓延和检测，从而可以检测到所有连通的肤色区域，并可以将所有连通的肤色区域中各个像素的像素值在当前帧图片的肤色查找表中对应的肤色值设为第一数值，从而确定当前帧图片的肤色查找表。
方式二：从当前帧图片的灰度图中获取人脸区域，计算人脸区域的直方图，并找到该直方图的近似谷点，通过该近似谷点对人脸区域中的肤色区域和非肤色区域进行划分，将人脸
区域中的肤色区域中各个像素的像素值在当前帧图片的肤色查找表中对应的肤色值设为第一数值，从而确定当前帧图片的肤色查找表。
203、终端根据当前帧图片的肤色查找表和当前帧图片之前的历史帧图片累计的肤色查找表，确定出目标肤色查找表。
204、终端利用目标肤色查找表对当前帧图片进行肤色检测。
具体的，目标肤色查找表可以通过对当前帧图片的肤色查找表和当前帧图片之前的历史帧图片累计的肤色查找表进行加权平均得到，例如，目标肤色查找表中的肤色值＝(1-ω)*历史帧图片累计的肤色查找表中的肤色值+ω*当前帧图片的肤色查找表中的肤色值，ω为加权系数，可取ω＝5％，从而得到的是肤色值连续的目标肤色查找表，实现了对当前帧图片的肤色查找表中的肤色值的自动学***均，目标肤色查找表中的肤色值则是连续的，范围是[0，255]。终端获取当前帧图片中每一个像素的像素值，从当前帧图片的目标肤色查找表中，查询每一个像素的像素值对应的肤色值，像素值对应的肤色值为255的像素确定为肤色，像素值对应的肤色值为0的像素确定为非肤色，而除0和255以外的其它肤色值，则是肤色值越大，置信度越高，对应的像素越接近于肤色，从而可以精确地检测出当前帧图片的肤色区域。可见，利用肤色值连续的肤色查找表对视频的每一帧图片进行肤色检测，可以实现对视频进行肤色检测时检测结果的连续。
本发明实施例中，终端获取进行肤色检测的视频的当前帧图片，对该当前帧图片进行人脸识别，以确定该当前帧图片的肤色查找表，并根据该当前帧图片的肤色查找表和该当前帧图片之前的历史帧图片累计的肤色查找表，确定出肤色值连续的目标肤色查找表，从而利用该目标肤色查找表对该当前帧图片进行肤色检测，从而在对视频进行肤色检测时，可以有效保证肤色检测结果的连续性。
请参阅图3，为本发明实施例提供的一种肤色检测方法的第二实施例流程示意图。本实施例中所描述的肤色检测方法，包括以下步骤：
301、终端获取进行肤色检测的视频的当前帧图片。
在执行步骤301之前，终端可以对两个模板图片集(即第一模板图片集集和第二模板图片集)分别进行肤色标定，以得到两张肤色查找表(即第一肤色查找表和第二肤色查找表)，第一肤色查找表和第二肤色查找表用于对视频的肤色检测。
假设第一肤色查找表对应第一模板图片集，第二肤色查找表对应第二模板图片集，终端对第一模板图片集和第二模板图片集分别进行肤色标定，以得到第一肤色查找表和第二肤色查找表的具体步骤可以如下：
针对第一模板图片集包括的每一张模板图片，通过选定框的方式选中模板图片中的肤色区域，如图4所示的矩形选定框，通过在模板图片中的不同区域利用矩形选定框进行框选以选中肤色区域(例如人脸、脖子、胳膊、手部、腿部等)，对于选定框选中的每一个图片区域，获取选定框选中的图片区域包括的像素总数和每一个像素的像素值，为便于对当前帧图片中的像素进行插值操作以得到肤色值连续的肤色检测结果，对选定框选中的图片区域包括的每
一个像素的像素值损失预设位数的精度，本实施例以损失3位精度为例，则像素值加4，再右移3位，统计选中的图片区域中每一个像素值对应的像素数，确定出第一模板图片集包括的所有模板图片中，对应的像素数大于或等于像素总数预设比例(例如1％)的第一目标像素值，可以理解的是，第一目标像素值具体包括多个像素值。将像素值为第一目标像素值的像素认为是肤色，将像素值为其它像素值的像素认为是非肤色，从而将第一肤色查找表中第一目标像素值对应的肤色值设为第一数值(即肤色)，将第一肤色查找表中除第一目标像素值之外的其它像素值对应的肤色值设为第二数值(即非肤色)，从而完成第一肤色查找表的确定。
在一些可行的实施方式中，可以先针对第一模板图片集包括的每一张模板图片的肤色查找表进行肤色值的设定，再将每一张模板图片的肤色查找表进行叠加即可得到第一肤色查找表，具体可以是：针对第一模板图片集包括的每一张模板图片，确定对应的像素数大于或等于像素总数预设比例(例如1％)的第一目标像素值，将模板图片的肤色查找表中第一目标像素值对应的肤色值设为第一数值，除第一目标像素值之外的其它像素值对应的肤色值设为第二数值，最后将每一张模板图片的肤色查找表进行叠加，可以是取并集的方式，即对于同一个像素值，只要在一张模板图片的肤色查找表中对于的肤色值为第一数值，则将第一模板图片集的第一肤色查找表中相应的肤色值设为第一数值，对于在第一模板图片集包括的所有模板图片的肤色查找表中对于的肤色值均为第二数值的像素值，则将第一模板图片集的第一肤色查找表中相应的肤色值设为第二数值，从而完成第一肤色查找表的确定。
采用与上述同样的方式可以完成第二模板图片集的第二肤色查找表的确定。
其中，第一模板图片集和第二模板图片集的区别在于：第一模板图片集包括的模板图片的数量大于第二模板图片集包括的模板图片的数量，即第一模板图片集的覆盖范围更广，相比于第二模板图片集，可以对更多的像素值进行肤色标定。
其中，第一模板图片集具体可以包括终端在特定拍照参数下拍得的本地图片，以及互联网上的图片，第二模板图片集具体可以只包括终端在特定拍照参数下拍得的本地图片。第一模板图片集对应的特定拍照参数可以是多色温、自动曝光(Automatic Exposure，AE)、自动白平衡(Automatic White Balance，AWB)等，以拍得各个色温下的图片，以不漏检为原则，第二模板图片集对应的特定拍照参数可以是正常色温、AE、AWB等，以拍得正常色温下的图片，以不误检为原则。
在一些可行的实施方式中，基于终端拍得的人物图片中，相比于人体的其它部位，人脸的肤色区域较大，代表性较强，第一模板图片集、第二模板图片集包括的模板图片可以是至少包括有人脸的图片。
302、终端对当前帧图片进行人脸识别，若识别到人脸，则根据第一模板图片集的第一肤色查找表，确定当前帧图片的人脸区域的肤色查找表。
具体实现中，终端如果从当前帧图片中识别到人脸，则根据第一肤色查找表，确定人脸区域的肤色查找表，包括：获取人脸区域中每一个像素的像素值，对每一个像素的像素值同样损失3位精度，从第一肤色查找表中，查询人脸区域中每一个像素损失3位精度后的像素值对应的肤色值，如果第一像素值对应的肤色值为第一数值，则损失3位精度后像素值为第一像素值的像素为肤色，进而将人脸区域的肤色查找表中第一像素值对应的肤色值也设为第一数值，对于损失3位精度后，对应的肤色值为第二数值的像素值，则将其在人脸区域的肤色查找表中对应的肤色值也设为第二数值，利用覆盖范围更广的第一肤色查找表先确定人脸
区域的肤色查找表，即确定人脸区域中的肤色区域，可以有效检测出人脸区域中的非肤色部分，例如眼睛、嘴唇、眼镜、眉毛等，同时，还可以避免人脸识别出现偏差(例如识别出的人脸区域比实际区域偏大)时导致的肤色误检。
303、终端根据人脸区域的肤色查找表和第二模板图片集的第二肤色查找表，确定当前帧图片的肤色查找表。
具体实现中，对确定出的人脸区域的肤色查找表和第二肤色查找表通过取并集的方式，确定当前帧图片的肤色查找表，包括：如果人脸区域的肤色查找表和第二肤色查找表中，同一个像素值对应的肤色值有一个为第一数值或者均为第一数值，则可以将该同一个像素值在当前帧图片的肤色查找表中对应的肤色值设为第一数值；如果人脸区域的肤色查找表和第二肤色查找表中，同一个像素值对应的肤色值均为第二数值，则将该同一个像素值在当前帧图片的肤色查找表中对应的肤色值设为第二数值，通过第一肤色查找表、第二肤色查找表和人脸识别确定当前帧图片的肤色查找表，可以大大减少漏检率和误检率。
在一些可行的实施方式中，终端如果从当前帧图片中没有识别到人脸，则可以直接根据覆盖范围较小的第二肤色查找表，确定当前帧图片的肤色查找表，包括：获取当前帧图片中每一个像素的像素值，从第二肤色查找表中，查询当前帧图片中每一个像素的像素值对应的肤色值，如果第二像素值对应的肤色值为第一数值，则像素值为第二像素值的像素为肤色，进而将当前帧图片的肤色查找表中第二像素值对应的肤色值也设为第一数值，对于对应的肤色值为第二数值的像素值，则将其在当前帧图片的肤色查找表中对应的肤色值也设为第二数值。第二肤色查找表是根据终端以不误检为原则，在拍照参数为正常色温、AE、AWB时拍得的第二模板图片集确定的，第二肤色查找表具有较好的肤色检测能力，可以使得在从当前帧图片中没有识别到人脸的情况下仍然具备肤色检测能力。
304、终端根据当前帧图片的肤色查找表和当前帧图片之前的历史帧图片累计的肤色查找表，确定出目标肤色查找表。
305、终端利用目标肤色查找表对当前帧图片进行肤色检测。
具体实现中，终端获取当前帧图片中每一个像素的像素值，对每一个像素的像素值同样损失3位精度，从目标肤色查找表中，查询每一个像素损失3位精度后的像素值对应的肤色值，从而可以精确地检测出当前帧图片的肤色区域。或者，终端也可以利用目标肤色查找表对当前帧图片中的每一个像素进行插值，以得到肤色值连续的肤色检测结果。
例如，对当前帧图片中的每一个像素进行四面体线性插值，以当前帧图片中的任意一个像素为例，四面体线性插值的步骤如下：分别在R/G/B三轴上以8为间隔单位，则整个颜色空间被切分为32768个均匀小立方体(立方体边长为8)，将每个均匀立方体按照特定的规则划分成除面重叠之外没有任何重叠的六个四面体，再确定像素所属的四面体，已知四面体四个顶点的坐标值，根据当前帧图片的肤色查找表可知四个顶点的肤色值，通过线性插值即可插出像素的肤色值。
306、终端对经过肤色检测的当前帧图片进行导向滤波，得到蒙版图片。
具体实现中，导向滤波有两个输入，一个是输入图p，一个是导向图I，有一个输出q。本实施例中，p为插值结果，I为输入图像(即当前帧图片)的灰度图或单通道图，q为优化后的肤色检测结果图。
其中，导向滤波是基于局部线性模型的，认为图像是一个二维函数，而且没法写出解析表达式，因此假设该函数的输入(导向图)与输出在一个窗口内满足线性关系如下：
对式(1)两边取梯度，当导向图I有梯度时，输出q也有类似的梯度，q能够保持导向图I的边界。对式(1)做线性回归，即希望拟合函数的输出值和真实值p之间的差距最小：
利用最小二乘法求出a和b：
具体求某一点的输出值时，只需要将所有包含改点的线性函数平均即可：
输出q的计算步骤可以如下：
meanI＝fmean(I)
meanp＝fmean(p)
corrI＝fmean(I*I)
corrIp＝fmean(I*p)
varI＝corrI–meanI*meanI
covIp＝corrIp–meanI*meanp
a＝covIp/(varI+e)
b＝meanp–a*meanI
meana＝fmean(a)
meanb＝fmean(b)
q＝meana*I+meanb
其中，fmean的功能是半径为20的窗口像素的平均值，e的作用是使除数不为0，e的
值越小越好。
307、终端对当前帧图片进行美化处理，得到经过美化的当前帧图片。
308、终端利用该蒙版图片将当前帧图片和经过美化的当前帧图片进行融合。
具体实现中，终端对当前帧图片加入美化处理效果，例如磨皮、美白等，再利用蒙版图片将当前帧图片和经过美化的当前帧图片进行融合(透明度混合)，即result＝(1-α)*image1+α*image2，其中，α为蒙版图片，image1为当前帧图片，image2为经过美化的当前帧图片，result为最终处理结果，从而得到具有肤色检测准确，消除视觉上的跳变现象，以及美颜效果良好的视频。
本发明实施例中，终端获取进行肤色检测的视频的当前帧图片，根据第一模板图片集的第一肤色查找表，确定当前帧图片的人脸区域的肤色查找表，根据人脸区域的肤色查找表和第二模板图片集的第二肤色查找表，确定当前帧图片的肤色查找表，根据当前帧图片的肤色查找表和当前帧图片之前的历史帧图片累计的肤色查找表，确定出肤色值连续的目标肤色查找表，利用目标肤色查找表对当前帧图片进行肤色检测，对经过肤色检测的当前帧图片进行导向滤波，得到蒙版图片，对当前帧图片进行美化处理，得到经过美化的当前帧图片，利用蒙版图片将当前帧图片和经过美化的当前帧图片进行融合，从而在对视频进行肤色检测时，可以有效保证肤色检测结果的连续性，再根据连续的肤色检测结果对视频进行美颜处理时可以消除视觉上的跳变现象，提供良好的视频美颜效果。
请参阅图5，为本发明实施例提供的一种肤色检测装置的结构示意图。本实施例中所描述的肤色检测装置，包括：
获取模块501，用于获取进行肤色检测的视频的当前帧图片。
确定模块502，用于对当前帧图片进行人脸识别，以确定当前帧图片的肤色查找表。
具体的，确定模块502若识别到人脸，则根据第一模板图片集的第一肤色查找表，确定当前帧图片的人脸区域的肤色查找表，再根据人脸区域的肤色查找表和第二模板图片集的第二肤色查找表，确定当前帧图片的肤色查找表。
其中，第一模板图片集包括的模板图片的数量大于第二模板图片集包括的模板图片的数量。
确定模块502若未识别到人脸，则根据第二肤色查找表，确定当前帧图片的肤色查找表。
确定模块502，还用于根据当前帧图片的肤色查找表和当前帧图片之前的历史帧图片累计的肤色查找表，确定出目标肤色查找表。
具体的，确定模块502对当前帧图片的肤色查找表和所述当前帧图片之前的历史帧图片累计的肤色查找表进行加权平均，得到目标肤色查找表。
其中，目标肤色查找表中的肤色值＝(1-ω)*历史帧图片累计的肤色查找表中的肤色值+ω*当前帧图片的肤色查找表中的肤色值，ω为加权系数，一般可取ω＝5％。
检测模块503，还用于利用目标肤色查找表对当前帧图片进行肤色检测。
可选的，该装置还包括：
滤波模块504，用于对经过肤色检测的当前帧图片进行导向滤波，得到蒙版图片。
美化模块505，用于对当前帧图片进行美化处理，得到经过美化的当前帧图片。
融合模块506，用于利用蒙版图片将当前帧图片和经过美化的当前帧图片进行融合。
可以理解的是，本实施例的肤色检测装置的各功能模块的功能可根据上述方法实施例中的方法具体实现，其具体实现过程可以参照上述方法实施例的相关描述，此处不再赘述。
本发明实施例中，终端获取进行肤色检测的视频的当前帧图片，对该当前帧图片进行人脸识别，以确定该当前帧图片的肤色查找表，并根据该当前帧图片的肤色查找表和该当前帧图片之前的历史帧图片累计的肤色查找表，确定出肤色值连续的目标肤色查找表，从而利用该目标肤色查找表对该当前帧图片进行肤色检测，从而在对视频进行肤色检测时，可以有效保证肤色检测结果的连续性。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。其中，所述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)或随机存取存储器(Random Access Memory，简称RAM)等。
以上所揭露的仅为本发明较佳实施例而已，当然不能以此来限定本发明之权利范围，因此依本发明权利要求所作的等同变化，仍属本发明所涵盖的范围。
Claims (13)
- 一种肤色检测方法，其特征在于，包括：获取进行肤色检测的视频的当前帧图片；对所述当前帧图片进行人脸识别，以确定所述当前帧图片的肤色查找表；根据所述当前帧图片的肤色查找表和所述当前帧图片之前的历史帧图片累计的肤色查找表，确定用于对所述当前帧图片进行肤色检测的目标肤色查找表；利用所述目标肤色查找表对所述当前帧图片进行肤色检测。
- 根据权利要求1所述的方法，其特征在于，所述根据所述当前帧图片的肤色查找表和所述当前帧图片之前的历史帧图片累计的肤色查找表，确定用于对所述当前帧图片进行肤色检测的目标肤色查找表，包括：对所述当前帧图片的肤色查找表和所述当前帧图片之前的历史帧图片累计的肤色查找表进行加权平均，得到用于对所述当前帧图片进行肤色检测的目标肤色查找表。
- 根据权利要求2所述的方法，其特征在于，所述目标肤色查找表中的肤色值＝(1-ω)*所述历史帧图片累计的肤色查找表中的肤色值+ω*所述当前帧图片的肤色查找表中的肤色值，所述ω为加权系数。
- 根据权利要求1-3中任一项所述的方法，其特征在于，所述对所述当前帧图片进行人脸识别，以确定所述当前帧图片的肤色查找表，包括：对所述当前帧图片进行人脸识别；若识别到人脸，则根据第一模板图片集的第一肤色查找表，确定所述当前帧图片的人脸区域的肤色查找表；根据所述人脸区域的肤色查找表和第二模板图片集的第二肤色查找表，确定所述当前帧图片的肤色查找表；其中，所述第一模板图片集包括的模板图片的数量大于所述第二模板图片集包括的模板图片的数量。
- 根据权利要求4所述的方法，其特征在于，所述方法还包括：若未识别到人脸，则根据所述第二肤色查找表，确定所述当前帧图片的肤色查找表。
- 根据权利要求1所述的方法，其特征在于，所述利用所述目标肤色查找表对所述当前帧图片进行肤色检测之后，所述方法还包括：对经过肤色检测的所述当前帧图片进行导向滤波，得到蒙版图片；对所述当前帧图片进行美化处理，得到经过美化的所述当前帧图片；利用所述蒙版图片将所述当前帧图片和经过美化的所述当前帧图片进行融合。
- 一种肤色检测装置，其特征在于，包括：获取模块，用于获取进行肤色检测的视频的当前帧图片；确定模块，用于对所述当前帧图片进行人脸识别，以确定所述当前帧图片的肤色查找表；所述确定模块，还用于根据所述当前帧图片的肤色查找表和所述当前帧图片之前的历史帧图片累计的肤色查找表，确定用于对所述当前帧图片进行肤色检测的目标肤色查找表；检测模块，还用于利用所述目标肤色查找表对所述当前帧图片进行肤色检测。
- 根据权利要求7所述的装置，其特征在于，所述确定模块具体用于：对所述当前帧图片的肤色查找表和所述当前帧图片之前的历史帧图片累计的肤色查找表进行加权平均，得到用于对所述当前帧图片进行肤色检测的目标肤色查找表。
- 根据权利要求8所述的装置，其特征在于，所述目标肤色查找表中的肤色值＝(1-ω)*所述历史帧图片累计的肤色查找表中的肤色值+ω*所述当前帧图片的肤色查找表中的肤色值，所述ω为加权系数。
- 根据权利要求7-9中任一项所述的装置，其特征在于，所述确定模块具体用于：对所述当前帧图片进行人脸识别；若识别到人脸，则根据第一模板图片集的第一肤色查找表，确定所述当前帧图片的人脸区域的肤色查找表；根据所述人脸区域的肤色查找表和第二模板图片集的第二肤色查找表，确定所述当前帧图片的肤色查找表；其中，所述第一模板图片集包括的模板图片的数量大于所述第二模板图片集包括的模板图片的数量。
- 根据权利要求10所述的装置，其特征在于，所述确定模块还具体用于：若未识别到人脸，则根据所述第二肤色查找表，确定所述当前帧图片的肤色查找表。
- 根据权利要求7所述的装置，其特征在于，所述装置还包括：滤波模块，用于对经过肤色检测的所述当前帧图片进行导向滤波，得到蒙版图片；美化模块，用于对所述当前帧图片进行美化处理，得到经过美化的所述当前帧图片；融合模块，用于利用所述蒙版图片将所述当前帧图片和经过美化的所述当前帧图片进行融合。
- 一种终端，其特征在于，所述终端包括：处理器和储存器，所述处理器和所述存储器通过总线连接，所述存储器存储有可执行程序代码，所述处理器用于调用所述可执行程序代码，执行如权利要求1-6中任一项所述的肤色检测方法。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610950691.0 | 2016-11-02 | ||
| CN201610950691.0A CN106570472B (zh) | 2016-11-02 | 2016-11-02 | 一种肤色检测方法、装置及终端 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2018082388A1 true WO2018082388A1 (zh) | 2018-05-11 |
Family
ID=58534924
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2017/099869 WO2018082388A1 (zh) | 2016-11-02 | 2017-08-31 | 一种肤色检测方法、装置及终端 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN106570472B (zh) |
| WO (1) | WO2018082388A1 (zh) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111124862A (zh) * | 2019-12-24 | 2020-05-08 | 北京安兔兔科技有限公司 | 智能设备性能测试方法、装置及智能设备 |
| CN111815653A (zh) * | 2020-07-08 | 2020-10-23 | 深圳市梦网视讯有限公司 | 一种人脸与身体肤色区域的分割方法、***和设备 |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106570472B (zh) * | 2016-11-02 | 2019-11-05 | 华为技术有限公司 | 一种肤色检测方法、装置及终端 |
| CN106570909B (zh) * | 2016-11-02 | 2020-01-17 | 华为技术有限公司 | 一种肤色检测方法、装置及终端 |
| CN107424117B (zh) * | 2017-07-17 | 2021-03-30 | Oppo广东移动通信有限公司 | 图像美颜方法、装置、计算机可读存储介质和计算机设备 |
| CN108765503B (zh) * | 2018-05-21 | 2020-11-13 | 深圳市梦网科技发展有限公司 | 一种肤色检测方法、装置及终端 |
| EP3816932B1 (en) * | 2018-07-16 | 2022-06-15 | Honor Device Co., Ltd. | Skin detection method and electronic device |
| CN109089158B (zh) * | 2018-07-24 | 2020-04-28 | 四川长虹电器股份有限公司 | 用于智能电视的人脸画质参数处理***及其实现方法 |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040017930A1 (en) * | 2002-07-19 | 2004-01-29 | Samsung Electronics Co., Ltd. | System and method for detecting and tracking a plurality of faces in real time by integrating visual ques |
| CN102968623A (zh) * | 2012-12-07 | 2013-03-13 | 上海电机学院 | 肤色检测***及方法 |
| CN103106386A (zh) * | 2011-11-10 | 2013-05-15 | 华为技术有限公司 | 动态自适应肤色分割方法和装置 |
| CN103455790A (zh) * | 2013-06-24 | 2013-12-18 | 厦门美图网科技有限公司 | 一种基于肤色模型的皮肤识别方法 |
| CN104392211A (zh) * | 2014-11-12 | 2015-03-04 | 厦门美图网科技有限公司 | 一种基于显著性检测的皮肤识别方法 |
| CN104598914A (zh) * | 2013-10-31 | 2015-05-06 | 展讯通信(天津)有限公司 | 一种肤色检测的方法及装置 |
| CN106570472A (zh) * | 2016-11-02 | 2017-04-19 | 华为技术有限公司 | 一种肤色检测方法、装置及终端 |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2006338377A (ja) * | 2005-06-02 | 2006-12-14 | Fujifilm Holdings Corp | 画像補正方法および装置並びにプログラム |
| CN101299242B (zh) * | 2008-06-13 | 2010-08-25 | 腾讯科技（深圳）有限公司 | 一种人体肤色检测中阈值的确定方法及装置 |
| CN101448085B (zh) * | 2008-12-26 | 2013-08-21 | 北京中星微电子有限公司 | 一种支持人脸检测的摄像处理方法和*** |
| CN105224917B (zh) * | 2015-09-10 | 2019-06-21 | 成都品果科技有限公司 | 一种利用颜色空间创建肤色概率图的方法和*** |
-
2016
- 2016-11-02 CN CN201610950691.0A patent/CN106570472B/zh active Active
-
2017
- 2017-08-31 WO PCT/CN2017/099869 patent/WO2018082388A1/zh active Application Filing
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040017930A1 (en) * | 2002-07-19 | 2004-01-29 | Samsung Electronics Co., Ltd. | System and method for detecting and tracking a plurality of faces in real time by integrating visual ques |
| CN103106386A (zh) * | 2011-11-10 | 2013-05-15 | 华为技术有限公司 | 动态自适应肤色分割方法和装置 |
| CN102968623A (zh) * | 2012-12-07 | 2013-03-13 | 上海电机学院 | 肤色检测***及方法 |
| CN103455790A (zh) * | 2013-06-24 | 2013-12-18 | 厦门美图网科技有限公司 | 一种基于肤色模型的皮肤识别方法 |
| CN104598914A (zh) * | 2013-10-31 | 2015-05-06 | 展讯通信(天津)有限公司 | 一种肤色检测的方法及装置 |
| CN104392211A (zh) * | 2014-11-12 | 2015-03-04 | 厦门美图网科技有限公司 | 一种基于显著性检测的皮肤识别方法 |
| CN106570472A (zh) * | 2016-11-02 | 2017-04-19 | 华为技术有限公司 | 一种肤色检测方法、装置及终端 |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111124862A (zh) * | 2019-12-24 | 2020-05-08 | 北京安兔兔科技有限公司 | 智能设备性能测试方法、装置及智能设备 |
| CN111124862B (zh) * | 2019-12-24 | 2024-01-30 | 北京安兔兔科技有限公司 | 智能设备性能测试方法、装置及智能设备 |
| CN111815653A (zh) * | 2020-07-08 | 2020-10-23 | 深圳市梦网视讯有限公司 | 一种人脸与身体肤色区域的分割方法、***和设备 |
| CN111815653B (zh) * | 2020-07-08 | 2024-01-30 | 深圳市梦网视讯有限公司 | 一种人脸与身体肤色区域的分割方法、***和设备 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106570472A (zh) | 2017-04-19 |
| CN106570472B (zh) | 2019-11-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2018082388A1 (zh) | 一种肤色检测方法、装置及终端 | |
| WO2018082389A1 (zh) | 一种肤色检测方法、装置及终端 | |
| JP7027537B2 (ja) | 画像処理方法および装置、電子機器、ならびにコンピュータ可読記憶媒体 | |
| Emberton et al. | Underwater image and video dehazing with pure haze region segmentation | |
| US9569854B2 (en) | Image processing method and apparatus | |
| US20200412940A1 (en) | Method and device for image processing, method for training object detection model | |
| CN107771336B (zh) | 基于颜色分布的图像中的特征检测和掩模 | |
| Lai et al. | Single-image dehazing via optimal transmission map under scene priors | |
| WO2021022983A1 (zh) | 图像处理方法和装置、电子设备、计算机可读存储介质 | |
| CN105243371A (zh) | 一种人脸美颜程度的检测方法、***及拍摄终端 | |
| WO2018068420A1 (zh) | 图像处理方法和装置 | |
| CN111402170B (zh) | 图像增强方法、装置、终端及计算机可读存储介质 | |
| CN106981078B (zh) | 视线校正方法、装置、智能会议终端及存储介质 | |
| US20140079319A1 (en) | Methods for enhancing images and apparatuses using the same | |
| CN109640066B (zh) | 高精度稠密深度图像的生成方法和装置 | |
| CN110781770B (zh) | 基于人脸识别的活体检测方法、装置及设备 | |
| CN107622480A (zh) | 一种Kinect深度图像增强方法 | |
| CN111127476A (zh) | 一种图像处理方法、装置、设备及存储介质 | |
| CN108447068A (zh) | 三元图自动生成方法及利用该三元图的前景提取方法 | |
| CN113301320B (zh) | 图像信息处理方法、装置和电子设备 | |
| WO2017173578A1 (zh) | 一种图像增强方法及装置 | |
| CN107564085B (zh) | 图像扭曲处理方法、装置、计算设备及计算机存储介质 | |
| KR20110021500A (ko) | 이동객체의 실시간 추적과 거리 측정 방법 및 그 장치 | |
| US20190205689A1 (en) | Method and device for processing image, electronic device and medium | |
| CN114372931A (zh) | 一种目标对象虚化方法、装置、存储介质及电子设备 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 17868191 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 17868191 Country of ref document: EP Kind code of ref document: A1 |