CN110062173A - Image processor and image processing method, equipment, storage medium and intelligent terminal - Google Patents
Image processor and image processing method, equipment, storage medium and intelligent terminal Download PDFInfo
- Publication number
- CN110062173A CN110062173A CN201910200117.7A CN201910200117A CN110062173A CN 110062173 A CN110062173 A CN 110062173A CN 201910200117 A CN201910200117 A CN 201910200117A CN 110062173 A CN110062173 A CN 110062173A
- Authority
- CN
- China
- Prior art keywords
- image
- network model
- noise
- exposure
- processor
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
- H04N23/70—Circuitry for compensating brightness variation in the scene
- H04N23/73—Circuitry for compensating brightness variation in the scene by influencing the exposure time
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N25/00—Circuitry of solid-state image sensors [SSIS]; Control thereof
- H04N25/60—Noise processing, e.g. detecting, correcting, reducing or removing noise
- H04N25/62—Detection or reduction of noise due to excess charges produced by the exposure, e.g. smear, blooming, ghost image, crosstalk or leakage between pixels
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Image Processing (AREA)
- Studio Devices (AREA)
Abstract
The invention discloses image processors and image processing method, equipment, storage medium and intelligent terminal.The image processor includes: reception device, is configured for receiving an at least frame image, and by a received at least frame image transmitting to selection device；Selection device is configured to first nerves network model, selects the highest frame image of aesthetic measure from being transferred from an at least frame image described in reception device, and by the frame image transmitting of selection to denoising device；And denoising device, it is configured to nervus opticus network model, noise reduction process is carried out to the frame image.By carrying out the selection of image using neural network model and carrying out noise reduction process, the moment form that high-speed moving object is simply and easily captured using the user of intelligent terminal (such as mobile phone, tablet computer etc.) can be allowed.
Description
Technical field
The present invention relates to image processing techniques more particularly to image processors and image processing method, equipment, storage medium
And intelligent terminal.
Background technique
Camera work is difficult to obtain good effect on intelligent terminal (such as mobile phone, tablet computer etc.) always when solidifying,
Reason essentially consists in: camera work is generally used for capturing the moment form of fast moving objects when solidifying, for example, the water to drip
Drop, the sport car of high-speed motion, stone pound spray splash in water etc., and therefore, it is necessary to high shutter speeds and extremely short exposure
Between light time, the picture of that " in a flash " can be just captured；And in the case where the time for exposure is extremely short, light-inletting quantity is insufficient, therefore logical
Often need for sensitivity (ISO) to be transferred to these noises high, will be unable to introduce a large amount of noise with avoiding at this time, and introduce
Noise reduction capability beyond image-signal processor (ISP), so that image quality substantially reduces.
Therefore, the solution for the image quality photographed when can effectively improve solidifying is needed.
Summary of the invention
One of in order to solve problem above, the present invention provides image processor as described below and image processing method,
Equipment, storage medium and intelligent terminal.
In brief, the present invention proposes that a kind of combine utilizes the selection processing of neural network model and coagulating for noise reduction process
When camera work, so as to allow the user using common intelligent terminal (such as mobile phone, tablet computer etc.) simply and easily to catch
Grasp the moment form of high-speed moving object.
By using image processing method or equipment of the invention, using only common smart machine (such as mobile phone, plate
Computer etc.) without using dedicated image picking-up apparatus (such as camera, video camera etc.), high-speed moving object can be captured
Moment form.
According to one embodiment of present invention, a kind of image processor is provided, described image processor includes: to receive dress
It sets, is configured for receiving an at least frame image, and by a received at least frame image transmitting to selection device；Selection dress
It sets, is configured to first nerves network model, select beauty from being transferred from an at least frame image described in reception device
The highest frame image of sight degree, and by the frame image transmitting of selection to denoising device；And denoising device, it is configured for
By nervus opticus network model, noise reduction process is carried out to the frame image.
Optionally, reception device received image be filming apparatus time for exposure reduce and sensitivity improve
In the case of the image that shoots.
Optionally, the first nerves network model is the recurrence convolutional neural networks model that a figure goes out into numerical value, is used
In the aesthetic measure marking to an at least frame image；And/or the nervus opticus network model is a figure into the U schemed out
Shape neural network model, for removing the noise in image.
Optionally, described image processor further includes first nerves network model training device, the first nerves network
Model training apparatus is configured for: the image that acquisition multiframe is shot when the time for exposure is reduced and sensitivity improves；
It manually gives a mark at least part in acquired image, to obtain the score value of respective image；And use is scored
Image and corresponding score value train first nerves network model.
Optionally, described image processor further includes nervus opticus network model training device, the nervus opticus network
Model training apparatus is configured for: when increasing the time for exposure and reducing sensitivity, being shot using specialized camera more
Frame low noise image, as supervision image；When the time for exposure is reduced and sensitivity improves, shot using intelligent terminal
Multiple image carrys out the noise parameter of estimating noise of input image, and according to the noise parameter of estimation, manually raw on the supervision image
At noise, noise image is obtained；Nervus opticus network model is trained using the noise image and the supervision image.
According to one embodiment of present invention, a kind of image processing method is provided, described image processing method includes: to receive
Step receives an at least frame image；Step is selected, by first nerves network model, from a received at least frame image
The middle highest frame image of selection aesthetic measure；And noise reduction step, by nervus opticus network model, to being selected in selection step
The frame image out carries out noise reduction process.
Optionally, the received image of receiving step institute is shot in the case where reducing the time for exposure and improving sensitivity
Image.
Optionally, the first nerves network model is the recurrence convolutional neural networks model that a figure goes out into numerical value, is used
In the aesthetic measure marking to an at least frame image；And/or the nervus opticus network model is a figure into the U schemed out
Shape neural network model, for removing the noise in image.
Optionally, train first nerves network model in the following manner: acquisition multiframe reduced and feels in the time for exposure
The image that luminosity is shot in the case of improving；Manually give a mark at least part in acquired image, it is corresponding to obtain
The score value of image；And first nerves network model is trained using the image and corresponding score value that are scored.
According to one embodiment of present invention, a kind of image processing equipment is provided, comprising: processor；And memory,
On be stored with executable code, when the executable code is executed by the processor, execute the processor and retouch above
One of method stated.
According to one embodiment of present invention, a kind of non-transitory machinable medium is provided, being stored thereon with can
Code is executed, when the executable code is executed by processor, the processor is made to execute one of method described above.
According to one embodiment of present invention, a kind of intelligent terminal is provided, comprising: filming apparatus, for shooting image；With
And one of image processing apparatus, including image processor recited above or image processing equipment recited above or on
Non-transitory machinable medium described in face, for handling the image that filming apparatus is shot.
The present invention can be allowed by carrying out the selection of image using neural network model and carrying out noise reduction process using general
The user of logical intelligent terminal (such as mobile phone, tablet computer etc.) simply and easily captures the moment form of high-speed moving object.
Detailed description of the invention
Disclosure illustrative embodiments are described in more detail in conjunction with the accompanying drawings, the disclosure above-mentioned and its
Its purpose, feature and advantage will be apparent, wherein in disclosure illustrative embodiments, identical appended drawing reference
Typically represent same parts.
Fig. 1 gives the schematic block diagram of the image processor of an exemplary embodiment according to the present invention.
Fig. 2 shows the effects handled by using image procossing scheme of the invention image.
Fig. 3 gives the schematic flow chart of the image processing method of an exemplary embodiment according to the present invention.
Fig. 4 gives the schematic block diagram of the image processing equipment of an exemplary embodiment according to the present invention.
Fig. 5 gives the schematic block diagram of the intelligent terminal of an exemplary embodiment according to the present invention.
Specific embodiment
The preferred embodiment of the disclosure is more fully described below with reference to accompanying drawings.Although showing the disclosure in attached drawing
Preferred embodiment, however, it is to be appreciated that may be realized in various forms the disclosure without the embodiment party that should be illustrated here
Formula is limited.On the contrary, these embodiments are provided so that this disclosure will be more thorough and complete, and can be by the disclosure
Range is completely communicated to those skilled in the art.What needs to be explained here is that number, serial number and attached drawing in the application
Mark it is merely for convenience description and occur, for step of the invention, sequence etc. be not limited in any way, unless
The execution that step has been explicitly pointed out in specification has specific sequencing.
Fig. 1 gives the schematic block diagram of the image processor of an exemplary embodiment according to the present invention.
As shown in Figure 1, the image processor 100 of an exemplary embodiment according to the present invention may include: to receive dress
Set 110, selection device 120, denoising device 130.
Firstly, reception device 110 receives filming apparatus (such as the camera of mobile phone or tablet computer by intelligent terminal
Deng) multiple image of shooting, and it is transferred to selection device 120.
Wherein, the multiple image of filming apparatus shooting for example can be the time for exposure is short, ISO high in the case where shoot
Multiple image.Further, this multiple image can be 32 frame images.It note that multiple image can be at least one image,
The present invention for multiple image quantity without additional limitation.
Here, for example, above-mentioned " time for exposure is short " also refers to, or even the filming apparatus of intelligent terminal can be reached
Minimum exposure time can be and be less than number centiseconds, as long as making to connect for example, for current common filming apparatus
In the photo of continuous shooting at least clearly.Even, the time for exposure for being as short as several signas can be used.
In addition, above-mentioned " ISO high " is also referred to, the ISO of filming apparatus can be adjusted to its maximum ISO value.Specifically
Ground, for example, the high ISO that the present invention uses can be 1600 or more, even higher than for current common filming apparatus
3200。
In addition, the reason of short, ISO high image, being explained herein as follows for multiple image for the time for exposure: due to
The light-inletting quantity of the aperture of filming apparatus is directly proportional to the time for exposure, also directly proportional to sensitivity (ISO), and if the time for exposure
Long, being easy capture, (time for exposure is longer, and the brightness of image is bigger, and noise is more less than clear moment picture.Especially shooting
When moving object, the time for exposure, the long object that will lead to movement generated smear), so when needing to improve ISO and reducing exposure
Between guarantee sufficient light-inletting quantity and clear moment picture (such as clear moment picture of moving object) can be captured.
Moreover, from the foregoing, it will be observed that the time for exposure, which shortens, can reduce the probability for taking blurred picture.
In addition, the time for exposure how short becomes and how high ISO become, there is relationship with ambient brightness.Exposure given above
The numerical value of time and ISO, for understanding, and should not be used as limitation of the invention only as example.
Selection device 120 can be configured to first nerves network model 125, from being transferred from reception device 110
Multiple image in select the highest frame image of aesthetic measure, and by the frame image transmitting of selection to denoising device 130.
Here, above-mentioned first nerves network model 125 can be the neural network model that a figure goes out into numerical value, example
Such as, it can be a convolutional neural networks model.Preferably, the network structure of first nerves network model 125 can be recurrence
The structure of convolutional neural networks model.
It can extract the feature in every frame image, analyze the characteristics of image of extraction, and export the frame image
Score, for example, score is higher, can to represent the frame image more beautiful.That is, the score of the output as first nerves network model
It can be used to indicate that the aesthetics of respective image.Here, aesthetics can indicate various by fuzziness, colour temperature, color matching, composition etc.
" good-looking " degree that factor influences.
Specifically, each to obtain for example, received multiple image can be made to pass sequentially through first nerves network model 125
The score of frame image, then selects a frame image of highest scoring, and is transmitted to denoising device 130.
Denoising device 130 can be configured to nervus opticus network model 135, select selection device and transmit
That frame image come carries out noise reduction process.
Here, nervus opticus network model 135 can be a figure into the neural network model schemed out, for example, it may be
One full convolutional neural networks model.Preferably, the structure of nervus opticus network model 135 can be U-shaped network structure.Second
Neural network model can remove the high intensity noise in high ISO image, obtain the muting clean image of a frame.
The selection device and denoising device each with neural network model through the invention, can effectively improve solidifying
When the image quality photographed.
Fig. 2 shows the effects handled by using image procossing scheme of the invention image.As shown in Fig. 2,
It is the image of common intelligent terminal (such as camera of regular handset etc.) shooting on the left of Fig. 2, is via this hair on the right side of Fig. 2
Bright image processor obtains a final image after handling the multiple image shot by common intelligent terminal.From Fig. 2
As can be seen that by using image procossing scheme of the invention, that is, pass through the first nerves network model 125 of selection device 120
Selection is carried out to multiple image and noise reduction is carried out to the image selected by the nervus opticus network model 135 of noise reduction transposition 130
Treated, and that frame image (image right of Fig. 2) is apparent, it appears that also more attractive.
The training data of first nerves network model 125 is generated, intelligent terminal, such as mobile phone can be used, is come big
Amount acquires the image of (being also possible to be continuously shot) the multiframe short exposure time shot under various scenes, high ISO, then leads to
It crosses artificial (such as can be by mark person) to give a mark to these images, the is trained with described image and numerical value of these marking
One neural network model.It note that the mark person of this paper can be the people of aesthstic grounding in basic skills.In addition, the standard of marking can wrap
The parameter related with the image quality of image such as fuzziness, colour temperature, color matching, composition is included, each image is determined by these parameters
For indicate image aesthetics score value.
As described above, here, for example, above-mentioned " time for exposure is short " also refers to, or even intelligent terminal can be reached
Filming apparatus minimum exposure time, for example, can be one less than hundreds of points for current common filming apparatus
Second, as long as making in the photo being continuously shot at least clearly.Even, the exposure for being as short as several signas can be used
Between light time.
In addition, above-mentioned " ISO high " is also referred to, the ISO of filming apparatus can be adjusted to its maximum ISO value.Specifically
Ground, for example, the high ISO that the present invention uses can be 1600 or more, even higher than for current common filming apparatus
3200。
Compared to it is traditional solidifying when the method for selection picture frame photographed, the selection of the invention using neural network model
The method of picture frame can relatively accurately choose suitable image, to facilitate subsequent processing.
The training data of nervus opticus network model is generated, specialized camera (such as single-lens reflex camera) can be used and come low
In the case of ISO, the long time for exposure (" long time for exposure and low ISO ") acquire some low noise images as supervision images,
Then intelligent terminal, such as mobile phone are used, shoots some images at specific environment (short exposure time, high ISO) to estimate to make an uproar
Sound parameter, and noise is manually generated on supervision image according to noise parameter and obtains noise image, schemed with noise image and supervision
As training nervus opticus network model.
Here, above-mentioned " low ISO " can refer to the minimum ISO value of specialized camera (for example, less than 100.Certainly, each
The ISO upper lower limit value of specialized camera is different), above-mentioned " long time for exposure " also refers to, in above-mentioned " low ISO " situation
Under, the time for exposure of bright photo can be taken.Here, it is noted that, since there are relationship in time for exposure and ISO, so here
" long time for exposure " not instead of regular time value may become with the variation of " low ISO " value of camera.But, due to
ISO value is low, so the time for exposure generally can be relatively longer, for instance it can be possible that the several seconds, so being referred to as " long exposure ".
Note that skilled person will appreciate that, " length " and " short " of this paper is all the in contrast reality that they are taken
Actual value is related with filming apparatus used.
Here, picture noise generally obeys the mixed distribution of Poisson distribution and Gaussian Profile, wherein the variance of Poisson distribution
It is expected that related with brightness and sensitivity, Gaussian Profile is square related with sensitivity, therefore, passes through the figure under acquisition specific environment
Picture, it is estimated that the distribution of noise parameter.Compared to it is traditional solidifying when the noise-reduction method photographed, of the invention uses nerve net
The noise-reduction method of network model can effectively promote noise reduction effect, for example, it is invisible high ISO bring noise can be dropped to vision
Degree, and this is that common noise reduction algorithm is not achieved.Because for high ISO bring noise, after common noise reduction algorithm noise reduction,
Still there is macroscopic noise, and image is easy to thicken.
The denoising device of embodiment according to the present invention allows to be imaged using high ISO visual without generating
Noise, do not obscured to allow to shoot the image that high speed object makes by short time exposure.
It note that the present invention for the first nerves network model for carrying out image selection and for removing noise
Which kind of neural network model two neural network models specifically use any restrictions are not added, as long as they can be realized required function
?.
By carrying out the selection of image using neural network model and carrying out noise reduction process, can allow using common intelligence
The user of terminal (such as mobile phone, tablet computer etc.) simply and easily captures the moment form of high-speed moving object.
It note that image processor of the invention and each device therein can be to be used with software, be also possible to
With hard-wired, there should not be any restrictions to this.
Optionally, the image processor of another exemplary embodiment according to the present invention can also include first nerves net
Network model training apparatus 124 and first nerves network model training device 134, as shown in Figure 1.
Wherein, first nerves network model training device 124 can be configured for acquisition multiframe in time for exposure reduction
And the image that sensitivity is shot in the case of improving；Manually give a mark at least part in acquired image, with
To the score value of respective image；And first nerves network model is trained using the image and corresponding score value that are scored.
Nervus opticus network model training device 134 can be configured for increasing the time for exposure and reducing sensitivity
In the case of, multiframe low noise image is shot using specialized camera, as supervision image；
When the time for exposure is reduced and sensitivity improves, image is estimated using intelligent terminal shoot multi-frame images
The noise parameter of noise, and according to the noise parameter of estimation, noise manually is generated on the supervision image, obtains noise pattern
Picture；Nervus opticus network model is trained using the noise image and the supervision image.
Here, for example, above-mentioned " increase the time for exposure and reduce sensitivity " refers to " long time for exposure and low
ISO”。
In addition, for example, above-mentioned " time for exposure is reduced and sensitivity improves " refers to " short exposure time and height
ISO”。
As described above, " short exposure time " also refers to, in addition can reach intelligent terminal filming apparatus it is most short
Time for exposure, for example, can be for current common filming apparatus and be less than number centisecond, as long as making to be continuously shot
Photo at least clearly.Even, the time for exposure for being as short as several signas can be used.
In addition, above-mentioned " high ISO " is also referred to, the ISO of filming apparatus can be adjusted to its maximum ISO value.Specifically
Ground, for example, the high ISO that the present invention uses can be 1600 or more, even higher than for current common filming apparatus
3200。
Fig. 3 gives the schematic flow chart of the image processing method of an exemplary embodiment according to the present invention.
In step S110 (receiving step), an at least frame image is received.
Here it is possible to image is directly received from the filming apparatus of intelligent terminal, it can also be from reception shooting elsewhere
Or pre-stored image, the present invention do not make additional limitation to this.
In step S120 (selection step), through first nerves network model 125 ', from a received at least frame image
Select the highest frame image of aesthetic measure.
In step S130 (noise reduction step), by nervus opticus network model 135 ', to what is selected in selection step S120
The frame image carries out noise reduction process.
Above-mentioned step S110-S130 is similar with above-mentioned device 110-130, and details are not described herein.
In addition, above-mentioned neural network model 125 ' and 135 ' is similar with above-mentioned neural network model 125 and 135,
This is also repeated no more.
It is alternatively possible to train first nerves network model in the following manner: acquisition multiframe reduces in the time for exposure
And the image that sensitivity is shot in the case of improving；Manually give a mark at least part in acquired image, with
To the score value of respective image；And first nerves network model is trained using the image and corresponding score value that are scored.
It is alternatively possible to train nervus opticus network model in the following manner: increasing the time for exposure and reducing sense
In the case of luminosity, multiframe low noise image is shot using specialized camera, as supervision image；The time for exposure reduce and it is photosensitive
In the case of degree improves, the noise parameter of estimating noise of input image, and making an uproar according to estimation are come using intelligent terminal shoot multi-frame images
Sound parameter manually generates noise on the supervision image, obtains noise image；Schemed using the noise image and the supervision
As training nervus opticus network model.
Here, for example, above-mentioned " increase the time for exposure and reduce sensitivity " refer to it is above-mentioned " when long exposure
Between and low ISO ".
As described above, " time for exposure reduce and sensitivity improve " here refer to it is above-mentioned " when short exposure
Between and high ISO ".
" short exposure time " also refers to, or even can reach the minimum exposure time of the filming apparatus of intelligent terminal,
For example, can be for current common filming apparatus and be less than number centisecond, as long as making in the photo being continuously shot extremely
It is rare clearly.Even, the time for exposure for being as short as several signas can be used.
In addition, above-mentioned " high ISO " is also referred to, the ISO of filming apparatus can be adjusted to its maximum ISO value.Specifically
Ground, for example, the high ISO that the present invention uses can be 1600 or more, even higher than for current common filming apparatus
3200。
By carrying out the selection of image using neural network model and carrying out noise reduction process, can allow using common intelligence
The user of terminal (such as mobile phone, tablet computer etc.) simply and easily captures the moment form of high-speed moving object.
Fig. 4 gives the schematic block diagram of the image processing equipment of an exemplary embodiment according to the present invention.
Referring to fig. 4, which includes memory 10 and processor 20.
Processor 20 can be the processor of a multicore, also may include multiple processors.In some embodiments, locate
Reason device 20 may include a general primary processor and one or more special coprocessors, such as graphics processor
(GPU), digital signal processor (DSP) etc..In some embodiments, the circuit realization of customization can be used in processor 20,
Such as application-specific IC (ASIC, Application Specific Integrated Circuit) or scene can
Programmed logic gate array (FPGA, Field Programmable Gate Arrays).
It is stored with executable code on memory 10, when the executable code is executed by the processor 20, makes institute
It states processor 20 and executes one of method described above.Wherein, memory 10 may include various types of storage units, such as
Installed System Memory, read-only memory (ROM) and permanent storage.Wherein, ROM can store processor 20 or computer
The static data or instruction that other devices need.Permanent storage can be read-write storage device.Permanently store dress
Set the non-volatile memory device that the instruction and data of storage will not be lost can be after computer circuit breaking.In some realities
It applies in mode, permanent storage device is used as permanent storage using mass storage device (such as magnetically or optically disk, flash memory).
In other embodiment, permanent storage device can be removable storage equipment (such as floppy disk, CD-ROM drive).In system
It deposits and can be read-write storage equipment or the read-write storage equipment of volatibility, such as dynamic random access memory.Installed System Memory
It can store the instruction and data that some or all processors need at runtime.In addition, memory 10 may include any
The combination of computer readable storage medium, including (DRAM, SRAM, SDRAM, flash memory can for various types of semiconductor memory chips
Program read-only memory), disk and/or CD can also use.In some embodiments, memory 10 may include readable
And/or the removable storage equipment write, such as laser disc (CD), read-only digital versatile disc (such as DVD-ROM, it is double
Layer DVD-ROM), read-only Blu-ray Disc, super disc density, flash card (such as SD card, min SD card, Micro-SD card etc.),
Magnetic floppy disc etc..Computer readable storage medium does not include carrier wave and the momentary electron signal by wirelessly or non-wirelessly transmitting.
In addition, being also implemented as a kind of computer program or computer program product, the meter according to the method for the present invention
Calculation machine program or computer program product include the calculating for executing the above steps limited in the above method of the invention
Machine program code instruction.
Alternatively, the present invention can also be embodied as a kind of (or the computer-readable storage of non-transitory machinable medium
Medium or machine readable storage medium), it is stored thereon with executable code (or computer program or computer instruction code),
When the executable code (or computer program or computer instruction code) by electronic equipment (or calculate equipment, server
Deng) processor execute when, so that the processor is executed each step according to the above method of the present invention.
Fig. 5 gives the schematic block diagram of the intelligent terminal of an exemplary embodiment according to the present invention.
As shown in figure 5, intelligent terminal 1000 may include: for shooting at the filming apparatus 1010 and image of image
Manage device 1020.
Wherein, filming apparatus 1010 can be can shooting image and be adapted for mount on appointing on intelligent terminal 1000
What shooting device, the present invention do not make any additional limitation to this.
In addition, image processing apparatus 1020 may include image processor 100 described above or described above
Image processing equipment 1 or non-transitory machinable medium described above etc., for being shot to filming apparatus 1010
Image handled, to obtain image quality better image.
Those skilled in the art will also understand is that, various illustrative logical blocks, dress in conjunction with described in disclosure herein
It sets, circuit and algorithm steps may be implemented as the combination of electronic hardware, computer software or both.
What flow chart and block diagram in attached drawing etc. showed the system and method for multiple embodiments according to the present invention can
The architecture, function and operation being able to achieve.In this regard, each box in flowchart or block diagram can represent a dress
It sets, a part of program segment or code, a part of described device, program segment or code includes one or more for realizing rule
The executable instruction of fixed logic function.It should also be noted that in some implementations as replacements, the function of being marked in box
It can also be occurred with being different from the sequence marked in attached drawing.For example, two continuous boxes can actually be substantially in parallel
It executes, they can also be executed in the opposite order sometimes, and this depends on the function involved.It is also noted that block diagram and/
Or the combination of each box in flow chart and the box in block diagram and or flow chart, can with execute as defined in function or
The dedicated hardware based system of operation is realized, or can be realized using a combination of dedicated hardware and computer instructions.
Various embodiments of the present invention are described above, above description is exemplary, and non-exclusive, and
It is not limited to disclosed each embodiment.Without departing from the scope and spirit of illustrated each embodiment, for this skill
Many modifications and changes are obvious for the those of ordinary skill in art field.The selection of term used herein, purport
In the principle, practical application or improvement to the technology in market for best explaining each embodiment, or make the art
Other those of ordinary skill can understand each embodiment disclosed herein.
Claims (13)
1. a kind of image processor, which is characterized in that described image processor includes:
Reception device is configured for receiving an at least frame image, and by a received at least frame image transmitting to selection
Device；
Selection device is configured to first nerves network model, from being transferred from an at least frame figure described in reception device
The highest frame image of aesthetic measure is selected as in, and by the frame image transmitting of selection to denoising device；And
Denoising device is configured to nervus opticus network model, carries out noise reduction process to the frame image.
2. image processor as described in claim 1, which is characterized in that the received image of reception device institute is in filming apparatus
Time for exposure reduce and the image that shoots in the case that sensitivity improves.
3. image processor as described in claim 1, which is characterized in that the first nerves network model is a figure into number
It is worth recurrence convolutional neural networks model out, for the aesthetic measure marking to an at least frame image；And/or
The nervus opticus network model is a figure into the U-shaped neural network model schemed out, for removing the noise in image.
4. image processor as described in claim 1, which is characterized in that described image processor further includes first nerves network
Model training apparatus, the first nerves network model training device are configured for:
The image that acquisition multiframe is shot when the time for exposure is reduced and sensitivity improves；
It manually gives a mark at least part in acquired image, to obtain the score value of respective image；And
First nerves network model is trained using the image and corresponding score value that are scored.
5. image processor as described in claim 1, which is characterized in that described image processor further includes nervus opticus network
Model training apparatus, the nervus opticus network model training device are configured for:
When increasing the time for exposure and reducing sensitivity, multiframe low noise image is shot using specialized camera, as prison
Superintend and direct image；
When the time for exposure is reduced and sensitivity improves, carry out estimating noise of input image using intelligent terminal shoot multi-frame images
Noise parameter manually generate noise on the supervision image, obtain noise image and according to the noise parameter of estimation；
Nervus opticus network model is trained using the noise image and the supervision image.
6. a kind of image processing method, which is characterized in that described image processing method includes:
Receiving step receives an at least frame image；
Step is selected, by first nerves network model, aesthetic measure highest is selected from a received at least frame image
A frame image；And
Noise reduction step carries out noise reduction process to the frame image selected in selection step by nervus opticus network model.
7. image processing method as claimed in claim 6, which is characterized in that the received image of receiving step institute is to reduce exposure
The image shot between light time and in the case where improving sensitivity.
8. image processing method as claimed in claim 6, which is characterized in that the first nerves network model be a figure into
The recurrence convolutional neural networks model that numerical value goes out, for the aesthetic measure marking to an at least frame image；And/or
The nervus opticus network model is a figure into the U-shaped neural network model schemed out, for removing the noise in image.
9. image processing method as claimed in claim 7, which is characterized in that train first nerves network mould in the following manner
Type:
The image that acquisition multiframe is shot when the time for exposure is reduced and sensitivity improves；
It manually gives a mark at least part in acquired image, to obtain the score value of respective image；And
First nerves network model is trained using the image and corresponding score value that are scored.
10. image processing method as claimed in claim 7, which is characterized in that train nervus opticus network in the following manner
Model:
In the case where increasing the time for exposure and reducing sensitivity, multiframe low noise image is shot using specialized camera, as
Supervise image；
When the time for exposure is reduced and sensitivity improves, carry out estimating noise of input image using intelligent terminal shoot multi-frame images
Noise parameter manually generate noise on the supervision image, obtain noise image and according to the noise parameter of estimation；
Nervus opticus network model is trained using the noise image and the supervision image.
11. a kind of image processing equipment, comprising:
Processor；And
Memory is stored thereon with executable code, when the executable code is executed by the processor, makes the processing
Device executes the method as described in any one of claim 6~10.
12. a kind of non-transitory machinable medium, is stored thereon with executable code, when the executable code is located
When managing device execution, the processor is made to execute the method as described in any one of claim 6~10.
13. a kind of intelligent terminal, comprising:
Filming apparatus, for shooting image；And
Image processing apparatus, including image processor or claim 11 described in any one in Claims 1 to 5
Non-transitory machinable medium described in the image processing equipment or claim 12, for being filled to shooting
The image for setting shooting is handled.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910200117.7A CN110062173A (en) | 2019-03-15 | 2019-03-15 | Image processor and image processing method, equipment, storage medium and intelligent terminal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910200117.7A CN110062173A (en) | 2019-03-15 | 2019-03-15 | Image processor and image processing method, equipment, storage medium and intelligent terminal |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110062173A true CN110062173A (en) | 2019-07-26 |
Family
ID=67316920
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910200117.7A Pending CN110062173A (en) | 2019-03-15 | 2019-03-15 | Image processor and image processing method, equipment, storage medium and intelligent terminal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110062173A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111612722A (en) * | 2020-05-26 | 2020-09-01 | 星际（重庆）智能装备技术研究院有限公司 | Low-illumination image processing method based on simplified Unet full-convolution neural network |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140355959A1 (en) * | 2013-05-29 | 2014-12-04 | Adobe Systems Incorporated | Multi-frame patch correspondence identification in video |
| CN106650737A (en) * | 2016-11-21 | 2017-05-10 | 中国科学院自动化研究所 | Image automatic cutting method |
| CN106875361A (en) * | 2017-02-17 | 2017-06-20 | 深圳市唯特视科技有限公司 | A kind of method that poisson noise is removed based on depth convolutional neural networks |
| CN107230192A (en) * | 2017-05-31 | 2017-10-03 | 广东欧珀移动通信有限公司 | Image processing method, device, computer-readable recording medium and mobile terminal |
| CN107610123A (en) * | 2017-10-11 | 2018-01-19 | 中共中央办公厅电子科技学院 | A kind of image aesthetic quality evaluation method based on depth convolutional neural networks |
| CN108234870A (en) * | 2017-12-27 | 2018-06-29 | 广东欧珀移动通信有限公司 | Image processing method, device, terminal and storage medium |
| CN108876735A (en) * | 2018-06-01 | 2018-11-23 | 武汉大学 | A kind of blind denoising method of true picture based on depth residual error network |
| CN108965731A (en) * | 2018-08-22 | 2018-12-07 | Oppo广东移动通信有限公司 | A kind of half-light image processing method and device, terminal, storage medium |
-
2019
- 2019-03-15 CN CN201910200117.7A patent/CN110062173A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140355959A1 (en) * | 2013-05-29 | 2014-12-04 | Adobe Systems Incorporated | Multi-frame patch correspondence identification in video |
| CN106650737A (en) * | 2016-11-21 | 2017-05-10 | 中国科学院自动化研究所 | Image automatic cutting method |
| CN106875361A (en) * | 2017-02-17 | 2017-06-20 | 深圳市唯特视科技有限公司 | A kind of method that poisson noise is removed based on depth convolutional neural networks |
| CN107230192A (en) * | 2017-05-31 | 2017-10-03 | 广东欧珀移动通信有限公司 | Image processing method, device, computer-readable recording medium and mobile terminal |
| CN107610123A (en) * | 2017-10-11 | 2018-01-19 | 中共中央办公厅电子科技学院 | A kind of image aesthetic quality evaluation method based on depth convolutional neural networks |
| CN108234870A (en) * | 2017-12-27 | 2018-06-29 | 广东欧珀移动通信有限公司 | Image processing method, device, terminal and storage medium |
| CN108876735A (en) * | 2018-06-01 | 2018-11-23 | 武汉大学 | A kind of blind denoising method of true picture based on depth residual error network |
| CN108965731A (en) * | 2018-08-22 | 2018-12-07 | Oppo广东移动通信有限公司 | A kind of half-light image processing method and device, terminal, storage medium |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111612722A (en) * | 2020-05-26 | 2020-09-01 | 星际（重庆）智能装备技术研究院有限公司 | Low-illumination image processing method based on simplified Unet full-convolution neural network |
| CN111612722B (en) * | 2020-05-26 | 2023-04-18 | 星际（重庆）智能装备技术研究院有限公司 | Low-illumination image processing method based on simplified Unet full-convolution neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107370958B (en) | Image blurs processing method, device and camera terminal | |
| CN107845128B (en) | Multi-exposure high-dynamic image reconstruction method with multi-scale detail fusion | |
| CN107220931B (en) | High dynamic range image reconstruction method based on gray level mapping | |
| KR101662846B1 (en) | Apparatus and method for generating bokeh in out-of-focus shooting | |
| US20180374261A1 (en) | Rendering glasses shadows | |
| US20110205236A1 (en) | Image processing apparatus and storage medium having stored therein an image processing program | |
| CN104469179A (en) | Method for combining dynamic pictures into mobile phone video | |
| CN108875619A (en) | Method for processing video frequency and device, electronic equipment, computer readable storage medium | |
| CN108111911B (en) | Video data real-time processing method and device based on self-adaptive tracking frame segmentation | |
| CN110324532A (en) | A kind of image weakening method, device, storage medium and electronic equipment | |
| JP6162345B2 (en) | Raw scene recognition that allows scene-dependent image modification before image recording or display | |
| CN107529091A (en) | Video clipping method and device | |
| WO2020057248A1 (en) | Image denoising method and apparatus, and device and storage medium | |
| CN109711246A (en) | A kind of dynamic object recognition methods, computer installation and readable storage medium storing program for executing | |
| CN108600729A (en) | Dynamic 3D models generating means and image generating method | |
| CN108391116A (en) | Total body scan unit based on 3D imaging technique and scan method | |
| CN110189354A (en) | Image processing method and image processor, image processing equipment and medium | |
| CN110062173A (en) | Image processor and image processing method, equipment, storage medium and intelligent terminal | |
| CN110516545A (en) | Model training, image processing method and equipment, image processor and medium | |
| CN111161685B (en) | Virtual reality display equipment and control method thereof | |
| Liu et al. | Bokeh effects based on stereo vision | |
| WO2021026848A1 (en) | Image processing method and device, and photographing apparatus, movable platform and storage medium | |
| KR20140072980A (en) | Apparatus and method for generating a High Dynamic Range image using single image | |
| CN108629329B (en) | Image processing method and device, electronic equipment and computer readable storage medium | |
| CN106231278A (en) | Method for processing video frequency and television system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190726 |
|
| RJ01 | Rejection of invention patent application after publication |