CN110929686A - Intrinsic image decomposition method, device, equipment and readable storage medium - Google Patents

Intrinsic image decomposition method, device, equipment and readable storage medium Download PDF

Info

Publication number
CN110929686A
CN110929686A CN201911256829.7A CN201911256829A CN110929686A CN 110929686 A CN110929686 A CN 110929686A CN 201911256829 A CN201911256829 A CN 201911256829A CN 110929686 A CN110929686 A CN 110929686A
Authority
CN
China
Prior art keywords
image
preset
intrinsic
target
intrinsic image
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
Application number
CN201911256829.7A
Other languages
Chinese (zh)
Inventor
饶军
龚文勇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Foshan Shining Pupil Technology Co.,Ltd.
Original Assignee
Shenzhen Sancheng Zhichuang Technology Co Ltd
Foshan Shining Pupil Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shenzhen Sancheng Zhichuang Technology Co Ltd, Foshan Shining Pupil Technology Co ltd filed Critical Shenzhen Sancheng Zhichuang Technology Co Ltd
Priority to CN201911256829.7A priority Critical patent/CN110929686A/en
Publication of CN110929686A publication Critical patent/CN110929686A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168Feature extraction; Face representation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16Human faces, e.g. facial parts, sketches or expressions

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 Analysis (AREA)

Abstract

The invention discloses an intrinsic image decomposition method, an intrinsic image decomposition device, intrinsic image decomposition equipment and a readable storage medium, wherein the intrinsic image decomposition method provides a preset constraint condition determined based on low-rank sparse properties of an illumination intrinsic image, and can better depict the properties of the intrinsic image and more completely separate coupled reflection and illumination target components compared with the existing constraint condition determined based on the sparse properties; the minimization objective function is optimized and solved by using an alternative direction multiplier method, compared with the existing intrinsic image decomposition method based on machine learning, the method is simpler and easier in calculation process, and the decomposition task can be completed without marking a large number of training samples; and the solving process based on the alternative direction multiplier method is a distributed computing process, so that the effect of real-time computing is realized.

Description

Intrinsic image decomposition method, device, equipment and readable storage medium
Technical Field
The present invention relates to the field of image processing technologies, and in particular, to an intrinsic image decomposition method, an intrinsic image decomposition device, an intrinsic image decomposition apparatus, and a readable storage medium.
Background
The intrinsic image is an illumination image and a reflection image obtained by decomposing an image. Wherein, the irradiation picture is a picture reflecting the illumination condition of the original picture; the reflection map is an image portion that can be maintained under varying illumination conditions, and is an original image from which highlights have been removed. The existing intrinsic image decomposition methods are mainly divided into two categories, one category is an intrinsic image decomposition method based on constraint optimization, the assumption of the method has limitation, the method is effective only under specific conditions, and the intrinsic image decomposition result with high separability under general conditions cannot be obtained; the other type is an intrinsic image decomposition method based on machine learning, which is limited by the difficulty in labeling the data set, so that a large number of image data sets with dense labels are difficult to obtain, and the intrinsic image decomposition result with high separability is also difficult to obtain. Due to the above problems, the two existing intrinsic image decomposition methods cannot obtain completely separated intrinsic images, thereby resulting in the technical problem that the intrinsic image decomposition method in the prior art is incomplete.
The above is only for the purpose of assisting understanding of the technical aspects of the present invention, and does not represent an admission that the above is prior art.
Disclosure of Invention
The invention mainly aims to provide an intrinsic image decomposition method, aiming at solving the technical problem that the intrinsic image decomposition is incomplete by the existing intrinsic image decomposition method.
To achieve the above object, the present invention provides an intrinsic image decomposition method applied to an intrinsic image decomposition device, the intrinsic image decomposition method comprising the steps of:
acquiring a preset constraint condition and a minimized objective function determined by a currently specified original image, wherein the preset constraint condition is determined according to the low-rank sparse property of an illumination intrinsic image;
and solving the minimized objective function based on an alternating direction multiplier method and the preset constraint condition to obtain an intrinsic image decomposition result of the original image.
Optionally, the step of solving the minimized objective function based on the alternative direction multiplier method and the preset constraint condition to obtain an intrinsic image decomposition result of the original image includes:
converting the minimized target function into an augmented Lagrangian function of the minimized target function based on a preset replacement rule;
generating a target iteration format of a target variable in the augmented Lagrangian function based on an alternating direction multiplier algorithm according to a preset conversion rule and the augmented Lagrangian function;
initializing the target iteration format, and performing iteration calculation on the target variable based on the formatted target iteration format;
and when the condition that the preset convergence condition is met is detected, taking a target variable value of the current iteration round number as a convergence target variable value to obtain the intrinsic image decomposition result in the convergence target variable value.
Optionally, the step of generating a target iteration format of the target variable in the augmented lagrangian function based on an alternating direction multiplier algorithm according to a preset conversion rule and the augmented lagrangian function includes:
generating an initial iteration format of the target variable based on the augmented Lagrangian function;
and converting the initial iteration format into the target iteration format according to a preset contraction operator rule and a singular value decomposition rule.
Optionally, before the step of taking the target variable value of the current iteration round as the convergence target variable value to obtain the eigen image decomposition result in the convergence target variable value when it is detected that the preset convergence condition is satisfied, the method further includes:
acquiring the change rate of the illumination target component in the target variable in the iteration process, and judging whether the change rate is smaller than a preset threshold value;
if the change rate is smaller than a preset threshold value, judging that a preset convergence condition is met;
and if the change rate is not less than a preset threshold value, judging that the preset convergence condition is not met.
Optionally, before the step of obtaining the preset constraint condition and the minimization objective function determined by the currently specified original image, the method further includes:
receiving an image sequence decomposition instruction sent by a user, and acquiring the number of images and an image pixel matrix in an original image determined based on the image sequence decomposition instruction;
and determining the minimization objective function based on a preset constraint condition, the number of the images and an image pixel matrix.
Optionally, the intrinsic image decomposition method further includes:
rank punishment is carried out on the minimization target function based on the nuclear norm of the illumination target component in the minimization target function.
Optionally, after the step of solving the minimization objective function based on the alternating direction multiplier method and the preset constraint condition to obtain an intrinsic image decomposition result of the original image, the method further includes:
when the original image is a face image, taking the illumination intrinsic image in the intrinsic image decomposition result as a face characteristic image;
and carrying out face recognition on the face characteristic image based on a preset image recognition algorithm.
Further, to achieve the above object, the present invention also provides an intrinsic image decomposition device comprising:
the target function determining module is used for acquiring a preset constraint condition and a minimized target function determined by the current specified original image, wherein the preset constraint condition is determined according to the low-rank sparse property of the illumination intrinsic image;
and the decomposition result acquisition module is used for solving the minimized objective function based on an alternating direction multiplier method and the preset constraint condition so as to acquire an intrinsic image decomposition result of the original image.
Further, to achieve the above object, the present invention also provides an intrinsic image decomposition device including: a memory, a processor and an intrinsic image decomposition program stored on the memory and executable on the processor, the intrinsic image decomposition program when executed by the processor implementing the steps of the intrinsic image decomposition method as described above.
Furthermore, to achieve the above object, the present invention also provides a computer readable storage medium having stored thereon an intrinsic image decomposition program which, when executed by a processor, implements the steps of the intrinsic image decomposition method as described above.
The invention provides an intrinsic image decomposition method, an intrinsic image decomposition device, intrinsic image decomposition equipment and a computer readable storage medium. The intrinsic image decomposition method comprises the steps of obtaining a preset constraint condition and a minimized target function determined by a currently specified original image, wherein the preset constraint condition is determined according to the low-rank sparse property of an illumination intrinsic image; and solving the minimized objective function based on an alternating direction multiplier method and the preset constraint condition to obtain an intrinsic image decomposition result of the original image. Through the method, the preset constraint condition determined based on the low-rank sparse property of the illumination intrinsic image is provided, and compared with the existing constraint condition determined based on the sparse property, the property of the intrinsic image can be better described, and the coupled reflection and illumination target component can be more completely separated; the minimization objective function is optimized and solved by using an alternative direction multiplier method, compared with the existing intrinsic image decomposition method based on machine learning, the method is simpler and easier in calculation process, and the decomposition task can be completed without marking a large number of training samples; and the solving process based on the alternative direction multiplier method is a distributed computing process, so that the effect of real-time computing is realized, and the technical problem that the intrinsic image is not completely decomposed by the existing intrinsic image decomposition method is solved.
Drawings
FIG. 1 is a schematic diagram of an apparatus architecture of a hardware operating environment according to an embodiment of the present invention;
FIG. 2 is a flowchart illustrating a first embodiment of the intrinsic image decomposition method according to the present invention;
FIG. 3 is a flowchart illustrating a second embodiment of the intrinsic image decomposition method according to the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
As shown in fig. 1, fig. 1 is a schematic terminal structure diagram of a hardware operating environment according to an embodiment of the present invention.
The terminal of the embodiment of the invention can be a PC, and can also be a mobile terminal device with a display function, such as a smart phone, a tablet computer, an electronic book reader, an MP3(Moving Picture Experts Group Audio Layer III, dynamic video Experts compress standard Audio Layer 3) player, an MP4(Moving Picture Experts Group Audio Layer IV, dynamic video Experts compress standard Audio Layer 3) player, a portable computer, and the like.
As shown in fig. 1, the terminal may include: a processor 1001, such as a CPU, a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. Wherein a communication bus 1002 is used to enable connective communication between these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1005 may be a high-speed RAM memory or a non-volatile memory (e.g., a magnetic disk memory). The memory 1005 may alternatively be a memory device separate from the processor 1001 described above.
Optionally, the terminal may further include a camera, a Radio Frequency (RF) circuit, a sensor, an audio circuit, a WiFi module, and the like. Such as light sensors, motion sensors, and other sensors. Specifically, the light sensor may include an ambient light sensor that may adjust the brightness of the display screen according to the brightness of ambient light, and a proximity sensor that may turn off the display screen and/or the backlight when the mobile terminal is moved to the ear. As one of the motion sensors, the gravity acceleration sensor can detect the magnitude of acceleration in each direction (generally, three axes), detect the magnitude and direction of gravity when the mobile terminal is stationary, and can be used for applications (such as horizontal and vertical screen switching, related games, magnetometer attitude calibration), vibration recognition related functions (such as pedometer and tapping) and the like for recognizing the attitude of the mobile terminal; of course, the mobile terminal may also be configured with other sensors such as a gyroscope, a barometer, a hygrometer, a thermometer, and an infrared sensor, which are not described herein again.
Those skilled in the art will appreciate that the terminal structure shown in fig. 1 is not intended to be limiting and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components.
As shown in fig. 1, a memory 1005, which is a kind of computer storage medium, may include therein an operating system, a network communication module, a user interface module, and an intrinsic image decomposition program.
In the terminal shown in fig. 1, the network interface 1004 is mainly used for connecting to a backend server and performing data communication with the backend server; the user interface 1003 is mainly used for connecting a client (user side) and performing data communication with the client; and the processor 1001 may be configured to invoke the intrinsic image decomposition program stored in the memory 1005 and perform the following operations:
acquiring a preset constraint condition and a minimized objective function determined by a currently specified original image, wherein the preset constraint condition is determined according to the low-rank sparse property of an illumination intrinsic image;
and solving the minimized objective function based on an alternating direction multiplier method and the preset constraint condition to obtain an intrinsic image decomposition result of the original image.
Further, the processor 1001 may call the intrinsic image decomposition program stored in the memory 1005, and also perform the following operations:
converting the minimized target function into an augmented Lagrangian function of the minimized target function based on a preset replacement rule;
generating a target iteration format of a target variable in the augmented Lagrangian function based on an alternating direction multiplier algorithm according to a preset conversion rule and the augmented Lagrangian function;
initializing the target iteration format, and performing iteration calculation on the target variable based on the formatted target iteration format;
and when the condition that the preset convergence condition is met is detected, taking a target variable value of the current iteration round number as a convergence target variable value to obtain the intrinsic image decomposition result in the convergence target variable value.
Further, the processor 1001 may call the intrinsic image decomposition program stored in the memory 1005, and also perform the following operations:
generating an initial iteration format of the target variable based on the augmented Lagrangian function;
and converting the initial iteration format into the target iteration format according to a preset contraction operator rule and a singular value decomposition rule.
Further, the processor 1001 may call the intrinsic image decomposition program stored in the memory 1005, and also perform the following operations:
acquiring the change rate of the illumination target component in the target variable in the iteration process, and judging whether the change rate is smaller than a preset threshold value;
if the change rate is smaller than a preset threshold value, judging that a preset convergence condition is met;
and if the change rate is not less than a preset threshold value, judging that the preset convergence condition is not met.
Further, the processor 1001 may call the intrinsic image decomposition program stored in the memory 1005, and also perform the following operations:
receiving an image sequence decomposition instruction sent by a user, and acquiring the number of images and an image pixel matrix in an original image determined based on the image sequence decomposition instruction;
and determining the minimization objective function based on a preset constraint condition, the number of the images and an image pixel matrix.
Further, the processor 1001 may call the intrinsic image decomposition program stored in the memory 1005, and also perform the following operations:
rank punishment is carried out on the minimization target function based on the nuclear norm of the illumination target component in the minimization target function.
Further, the processor 1001 may call the intrinsic image decomposition program stored in the memory 1005, and also perform the following operations:
when the original image is a face image, taking the illumination intrinsic image in the intrinsic image decomposition result as a face characteristic image;
and carrying out face recognition on the face characteristic image based on a preset image recognition algorithm.
Based on the above hardware structure, various embodiments of the intrinsic image decomposition method of the present invention are proposed.
Referring to fig. 2, fig. 2 is a flowchart illustrating a first embodiment of the eigen-image decomposition method.
A first embodiment of the present invention provides an intrinsic image decomposition method including the steps of:
the intrinsic image is an illumination image and a reflection image obtained by decomposing an image. Wherein, the irradiation picture is a picture reflecting the illumination condition of the original picture; the reflection map is an image portion that can be maintained under varying illumination conditions, and is an original image from which highlights have been removed.
Intrinsic decomposition assumes that the objects are perfectly scattering and is based on Retinex theory. Single intrinsic image decomposition one picture I is decomposed into: i ═ RL, i.e., the dot product of the reflectance eigenmap R of the picture and the illumination map L of the picture. The intrinsic reflectivity graph R is only related to the material and color of the object, and is not related to the intrinsic reflectivity of the object in the surrounding environment; the illumination environment interacts with the object geometry to form a luminance eigen-map L. In brief, the reflectance eigenmap R and the illumination eigenmap L reflect two mutually independent information, i.e., the color and the illumination of the object itself, respectively. For convenience of calculation, we generally log both sides of formula I ═ RL simultaneously, and multiply as an addition, i.e.: u + v, where f is log (i), u is log (r), and v is log (l). V can be calculated from f-u as long as u can be calculated. And the formula f u + v is a ill-conditioned problem, i.e. we know only f, but to calculate u and v, the number of unknowns is twice the number of known quantities. For pathological problems, it is common practice to add appropriate constraints or a priori knowledge to solve such problems. The existing intrinsic image decomposition methods are mainly divided into two categories, one category is an intrinsic image decomposition method based on constraint optimization, the assumption of the method has limitation, the method is effective only under specific conditions, and the intrinsic image decomposition result with high separability under general conditions cannot be obtained; the other type is an intrinsic image decomposition method based on machine learning, which is limited by the difficulty in labeling the data set, so that a large number of image data sets with dense labels are difficult to obtain, and the intrinsic image decomposition result with high separability is also difficult to obtain. Due to the above problems, the two existing intrinsic image decomposition methods cannot obtain completely separated intrinsic images, thereby resulting in the technical problem that the intrinsic image decomposition method in the prior art is incomplete.
In the embodiment, to solve the above problems, the invention provides an intrinsic image decomposition method, that is, a preset constraint condition determined based on the low-rank sparse property of an illumination intrinsic image is provided, and compared with the existing constraint condition determined based on the sparse property, the method can better depict the property of the intrinsic image, and can more completely separate the coupled reflection from the illumination target component; the minimization objective function is optimized and solved by using an alternative direction multiplier method, compared with the existing intrinsic image decomposition method based on machine learning, the method is simpler and easier in calculation process, and the decomposition task can be completed without marking a large number of training samples; and the solving process based on the alternative direction multiplier method is a distributed computing process, so that the effect of real-time computing is realized, and the technical problem that the intrinsic image is not completely decomposed by the existing intrinsic image decomposition method is solved. The intrinsic image decomposition method is applied to the terminal.
Step S10, acquiring a preset constraint condition and a minimized target function determined by the current specified original image, wherein the preset constraint condition is determined according to the low-rank sparse property of the illumination intrinsic image;
the preset constraint condition is a specific constraint condition for the scheme, which is determined in advance according to sparsity and low rank of the illumination intrinsic image. The original image is a plurality of images shot under the same visual angle and different lighting conditions of the same object, and the number of the original images is not limited in the embodiment.
In this embodiment, the terminal may receive an original image that is currently imported by a user or automatically acquired according to a preset program and currently needs to be subjected to eigen image decomposition. The number of the original images is determined according to specific situations, and is usually a group of image sequences including a plurality of images. And the terminal determines a minimized objective function to be solved according to a preset specific constraint condition for embodying low rank and sparsity and the current original image to be decomposed.
Specifically, a current set of pixel matrices Ii (i ═ 1, 2., n) including n well-aligned image sequences is set, the illumination target component is set to U, and the reflection target component is set to V. The group of image sequences all come from the same scene, so the reflection target components V of the n images in the group of image sequences are the same, and V may be set as [ V, V,.., V ]; while the illumination of the n images is different. Therefore, the reflection target components U are different from each other, and U ═ U1, U2. In a logarithmic sense, let v ui-fi, f be an M × n matrix, where M denotes that each picture has M pixels. Where fi ═ log (ii). f ═ f1, f 2.., fn ], the minimization objective function determined based on the above-mentioned set and preset constraints exhibiting low-rank sparse characteristics is:
Figure BDA0002309270660000091
wherein the content of the first and second substances,
Figure BDA0002309270660000092
d is a discrete format of gradient operatorA forward difference matrix of
Figure BDA0002309270660000093
Wherein
Figure BDA0002309270660000094
Is a kronecker product, and has
Figure BDA0002309270660000095
And E is defined as
Figure BDA0002309270660000096
And step S20, solving the minimization objective function based on an alternating direction multiplier method and the preset constraint condition to obtain an intrinsic image decomposition result of the original image.
In this embodiment, the terminal performs distributed iterative computation and solution on the minimized objective function based on an Alternating Direction multiplier (ADMM) and a preset constraint condition based on low rank and sparsity. It should be noted that since ui and v are derived from variable separation, the minimization constraint optimization problem of minimizing the objective function can be solved by using the ADMM. Specifically, the formula (1), that is, the minimized objective function, may be converted into the corresponding augmented lagrangian function, and then distributed iterative solution is performed on each objective variable in the obtained augmented lagrangian function based on the ADMM to obtain values of U and V in the objective variables when convergence is determined, where U and V are also intrinsic image decomposition results of the original image sequence.
The invention provides an intrinsic image decomposition method. The intrinsic image decomposition method comprises the steps of obtaining a preset constraint condition and a minimized target function determined by a currently specified original image, wherein the preset constraint condition is determined according to the low-rank sparse property of an illumination intrinsic image; and solving the minimized objective function based on an alternating direction multiplier method and the preset constraint condition to obtain an intrinsic image decomposition result of the original image. Through the method, the preset constraint condition determined based on the low-rank sparse property of the illumination intrinsic image is provided, and compared with the existing constraint condition determined based on the sparse property, the property of the intrinsic image can be better described, and the coupled reflection and illumination target component can be more completely separated; the minimization objective function is optimized and solved by using an alternative direction multiplier method, compared with the existing intrinsic image decomposition method based on machine learning, the method is simpler and easier in calculation process, and the decomposition task can be completed without marking a large number of training samples; and the solving process based on the alternative direction multiplier method is a distributed computing process, so that the effect of real-time computing is realized, and the technical problem that the intrinsic image is not completely decomposed by the existing intrinsic image decomposition method is solved.
Referring to fig. 3, fig. 3 is a flowchart illustrating a second embodiment of the intrinsic image decomposition method according to the present invention.
Based on the first embodiment shown in fig. 2 described above, in the present embodiment, step S20 includes:
step S21, converting the minimized objective function into an augmented Lagrangian function of the minimized objective function based on a preset replacement rule;
the preset replacement rule is a generation rule of replacement variables required for generating the augmented Lagrange function of the minimized objective function.
In this embodiment, for convenience of calculation, the replacement variable constituting the augmented lagrangian function corresponding to the minimized target function may be determined based on a preset replacement rule, and the augmented lagrangian function corresponding to the illumination target component, the reflection target component, and the replacement variable may be generated. Specifically, based on a preset replacement rule, let W ═ DU and a ═ U, where W and a are replacement variables. The augmented Lagrangian function corresponding to equation (1) in the first embodiment is
Figure BDA0002309270660000101
Where y1, y2, and y3 are lagrange multipliers.
Step S12, generating a target iteration format of a target variable in the augmented Lagrangian function based on an alternating direction multiplier algorithm according to a preset conversion rule and the augmented Lagrangian function;
wherein the target variables include at least an illumination target component U, a reflection target component V, and a replacement variable W, A. The preset conversion rule is a rule for converting the implicit iteration format of the target variable into the explicit iteration format.
In this embodiment, according to the principle of the ADMM optimization algorithm, the implicit iteration format of each target variable in the augmented lagrangian function corresponding to the minimized target function is obtained, and for facilitating subsequent iterative computation, the implicit iteration format of each target variable is converted into the explicit iteration format based on a preset conversion rule. Specifically, according to the formula (2) in the embodiment of step S21, the substitution variable W, A can be derived, and the implicit iteration format of the illumination target component U and the reflection target component V is as follows:
Figure BDA0002309270660000111
Figure BDA0002309270660000112
Figure BDA0002309270660000113
Figure BDA0002309270660000114
wherein c1, c2 and c3 are preset designated parameters, which is not limited in this embodiment. And transforming the implicit iterative formats of W and V according to the definition of a contraction operator, and performing singular value decomposition on the replacement variable A to obtain the respective explicit iterative formats of W, A and V as follows:
Figure BDA0002309270660000115
Figure BDA0002309270660000116
Figure BDA0002309270660000117
wherein
Figure BDA0002309270660000118
To represent
Figure BDA0002309270660000119
Column i.
Step S13, initializing the target iteration format, and performing iteration calculation on the target variable based on the formatted target iteration format;
in this embodiment, the converted display iteration format is initialized according to a preset initial value, so as to perform iterative computation according to the formatted target variable. Specifically, the initial values of the lagrange multipliers y1, y2, and y3 in equation (2) are set to
Figure BDA00023092706600001110
The initial value of the reflection target component V is set to
Figure BDA00023092706600001111
The initial value of the replacement variable W is set to 0. And ui is v + fi. And initializing the corresponding variable based on the initial value.
And step S14, when the preset convergence condition is met, taking the target variable value of the current iteration round number as a convergence target variable value to obtain the intrinsic image decomposition result in the convergence target variable value.
The preset convergence condition is that the change rate of the reflection target component is lower than a preset threshold value or the current iteration frequency reaches a preset maximum iteration frequency.
In this embodiment, when detecting that the current iterative computation satisfies the preset convergence condition, the terminal stops the current iterative computation, and obtains the illumination target component U and the reflection target component V obtained at the current iteration round number. At this time, the value of the illumination target component U and the value of the reflection target component V are the decomposition results of the current original image decomposition task. That is, the value of the illumination target component U at this time is the illumination intrinsic image of the original image, and the value of the illumination target component U at this time is the reflection intrinsic image of the original image.
Further, not shown in the figure, in the present embodiment, the step S22 includes:
step a, generating an initial iteration format of the target variable based on the augmented Lagrange function;
wherein, the initial iteration format is an implicit iteration format.
In the present embodiment, the substitution variable W, A can be derived from equation (2) in the embodiment of step S21, and the implicit iteration format of the illumination target component U and the reflection target component V is as follows:
Figure BDA0002309270660000121
Figure BDA0002309270660000122
Figure BDA0002309270660000123
Figure BDA0002309270660000124
wherein c1, c2 and c3 are preset designated parameters, which is not limited in this embodiment. In addition, the iterative format of the direct lagrangian operators y1, y2, y3 is as follows:
Figure BDA0002309270660000125
Figure BDA0002309270660000126
Figure BDA0002309270660000127
and b, converting the initial iteration format into the target iteration format according to a preset contraction operator rule and a singular value decomposition rule.
In the present embodiment, for the implicit iterative format of the replacement variable W and the reflected illumination component V, according to the definition of the contraction operator, for γ equal to 1,2, …, M, there is
Figure BDA0002309270660000128
Figure BDA0002309270660000129
Wherein the content of the first and second substances,
Figure BDA0002309270660000131
the explicit iterative format of the replacement variable W and the reflected illumination component V can be written as:
Figure BDA0002309270660000132
Figure BDA0002309270660000133
and because the replacing variable a relates to the kernel norm, Singular Value Decomposition (SVD) is required to be performed on the implicit iterative format of a. Based on in SVD
Figure BDA0002309270660000134
The explicit iteration format of the replacement variable a can be written as:
Figure BDA0002309270660000135
further, not shown, step S24 includes:
step c, acquiring the change rate of the illumination target component in the target variable in the iteration process, and judging whether the change rate is smaller than a preset threshold value;
the preset threshold is a numerical value for defining the change rate of the illumination target component in the iterative computation process, and the threshold can be flexibly set according to the actual situation, which is not limited in this embodiment.
In this embodiment, in each iteration, the terminal obtains the value of the illumination target component V in each iteration, calculates the change rate of V in each iteration, and compares the change rate with a preset change rate threshold. In addition, the convergence condition for the iterative process in the present invention may also be a preset maximum number of iterations. For example, the maximum iteration number may be preset to be 500, and if the change rate of the current illumination target component V is not less than the preset threshold value, but the terminal detects that the current iteration number is 500, it is determined that the preset convergence condition is reached currently, and the iterative computation is stopped.
D, if the change rate is smaller than a preset threshold value, judging that a preset convergence condition is met;
in this embodiment, if the terminal detects that the change rate of the illumination target component U of the current iteration round number is smaller than the preset threshold, it may be determined that the current augmented lagrangian function performing distributed iterative computation based on the ADMM optimization algorithm reaches convergence.
And e, if the change rate is not less than a preset threshold value, judging that the preset convergence condition is not met.
In this embodiment, if the terminal detects that the change rate of the illumination target component U of the current iteration round number is greater than or equal to the preset threshold, it may be determined that the current augmented lagrangian function performing distributed iterative computation based on the ADMM optimization algorithm does not reach convergence, and the iterative computation is continued until it is detected that the preset convergence condition is met, and the iterative computation is stopped.
The invention provides an intrinsic image decomposition method. The intrinsic image decomposition method further converts the minimized target function into an augmented Lagrangian function corresponding to the minimized target function by using a preset contraction operator and a singular value decomposition rule as preset replacement rules, and provides a precondition for the follow-up iterative computation based on ADMM; by respectively carrying out iterative computation on each target component in the augmented Lagrange function and acquiring an original image decomposition result when the function is converged, distributed computation can be completed without applying a machine learning algorithm, the computation process is simple and easy to implement, and the computation burden of the terminal is reduced; by comparing the change rate of the illumination component in the iterative calculation process with the preset threshold value, the convergence judgment process is simple and easy to implement, and the practicability of the method is improved.
In the drawings, a third embodiment of the intrinsic image decomposition method according to the present invention is proposed based on the first embodiment shown in fig. 2. In this embodiment, before step S10, the method further includes:
step f, receiving an image sequence decomposition instruction sent by a user, and acquiring the number of images and an image pixel matrix in the original image determined based on the image sequence decomposition instruction;
in this embodiment, the user selects the original image sequence that needs to be subjected to the eigen image decomposition at present, and clicks a button that is currently displayed by the terminal and used for creating the eigen image decomposition task. The terminal receives the image sequence decomposition instruction currently sent by the user, creates an image sequence decomposition task, obtains the number of images of the original image sequence in the image sequence decomposition instruction, and stores each image in the sequence in the form of a pixel matrix.
And g, determining the minimization objective function based on a preset constraint condition, the image number and the image pixel matrix.
In this embodiment, the terminal determines a minimization objective function for obtaining an eigen-image decomposition result corresponding to an image sequence according to a preset constraint condition determined by a low-rank sparse characteristic of an illumination eigen-image, the number of images in the image sequence to be decomposed in a current eigen-image decomposition task, and a pixel matrix of each image.
Further, in this embodiment, the intrinsic image decomposition method of the present invention further includes:
and h, carrying out rank punishment on the minimization target function based on the nuclear norm of the illumination target component in the minimization target function.
In this embodiment, the minimization objective function is expressed by the formula (1). Wherein | U ] of the third term*Represents the nuclear norm of the illumination target component U, and the term is used for carrying out rank punishment on the minimization target function represented by the formula (1) so as to embody low rank property.
Further, in this embodiment, after step S20, the method further includes:
step i, when the original image is a face image, taking an illumination intrinsic image in the intrinsic image decomposition result as a face characteristic image;
in this embodiment, the original image in the current eigen-image decomposition task is a plurality of images of the human face under different illumination conditions at the same time. And the terminal decomposes the intrinsic images according to the steps to obtain corresponding illumination intrinsic images and reflection intrinsic images, and takes the illumination intrinsic images as the face characteristic images.
And j, carrying out face recognition on the face characteristic image based on a preset image recognition algorithm.
The preset image recognition algorithm is an algorithm that can be used for face image recognition, and may be a Support Vector Machine (SVM), a K-nearest neighbor (kNN, K-nearest neighbor) classification algorithm, and the like.
In this embodiment, the terminal performs face recognition on the face feature image based on a preset image recognition algorithm, so that a more accurate recognition effect can be obtained. In addition, the intrinsic image decomposition result obtained by the method can also be used in the directions of image re-colorization, illumination migration of the human face image and the like, so that the method has better effect compared with the prior art.
The invention provides an intrinsic image decomposition method. The intrinsic image decomposition method further comprises the steps of obtaining the image number and the image pixel matrix in the image sequence decomposition instruction sent by the user, so that the minimization target function required by intrinsic image decomposition can be determined according to simple operation of the user; the kernel norm of the illumination target component is set in the minimized target function, so that rank punishment is carried out on the minimized target function, low rank performance is embodied, and the illumination target component and the reflection target component of the original image are separated more thoroughly; by carrying out face recognition on the basis of the invention, a recognition result with higher accuracy can be obtained.
The invention also provides an intrinsic image decomposition device.
The intrinsic image decomposition device includes:
the target function determining module is used for acquiring a preset constraint condition and a minimized target function determined by the current specified original image, wherein the preset constraint condition is determined according to the low-rank sparse property of the illumination intrinsic image;
and the decomposition result acquisition module is used for solving the minimized objective function based on an alternating direction multiplier method and the preset constraint condition so as to acquire an intrinsic image decomposition result of the original image.
The invention also provides intrinsic image decomposition equipment.
The intrinsic image decomposition device comprises a processor, a memory and an intrinsic image decomposition program stored on the memory and executable on the processor, wherein the intrinsic image decomposition program, when executed by the processor, implements the steps of the intrinsic image decomposition method as described above.
The method implemented when the intrinsic image decomposition program is executed may refer to various embodiments of the intrinsic image decomposition method of the present invention, and details thereof are not repeated herein.
The invention also provides a computer readable storage medium.
The computer readable storage medium of the present invention has stored thereon an intrinsic image decomposition program which, when executed by a processor, implements the steps of the intrinsic image decomposition method as described above.
The method implemented when the intrinsic image decomposition program is executed may refer to various embodiments of the intrinsic image decomposition method of the present invention, and details thereof are not repeated herein.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium (e.g., ROM/RAM, magnetic disk, optical disk) as described above and includes instructions for enabling a terminal device (e.g., a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims (10)

1. An intrinsic image decomposition method, comprising:
acquiring a preset constraint condition and a minimized objective function determined by a currently specified original image, wherein the preset constraint condition is determined according to the low-rank sparse property of an illumination intrinsic image;
and solving the minimized objective function based on an alternating direction multiplier method and the preset constraint condition to obtain an intrinsic image decomposition result of the original image.
2. The eigen-image decomposition method as claimed in claim 1, wherein the step of solving the minimized objective function based on the alternating direction multiplier method and the preset constraint condition to obtain the eigen-image decomposition result of the original image comprises:
converting the minimized target function into an augmented Lagrangian function of the minimized target function based on a preset replacement rule;
generating a target iteration format of a target variable in the augmented Lagrangian function based on an alternating direction multiplier algorithm according to a preset conversion rule and the augmented Lagrangian function;
initializing the target iteration format, and performing iteration calculation on the target variable based on the formatted target iteration format;
and when the condition that the preset convergence condition is met is detected, taking a target variable value of the current iteration round number as a convergence target variable value to obtain the intrinsic image decomposition result in the convergence target variable value.
3. The intrinsic image decomposition method according to claim 2, wherein the step of generating a target iteration format of a target variable in the augmented lagrangian function based on an alternating direction multiplier algorithm according to a preset transformation rule and the augmented lagrangian function comprises:
generating an initial iteration format of the target variable based on the augmented Lagrangian function;
and converting the initial iteration format into the target iteration format according to a preset contraction operator rule and a singular value decomposition rule.
4. The eigen-image decomposition method as claimed in claim 2, wherein before the step of taking the target variable value of the current iteration round as the convergence target variable value to obtain the eigen-image decomposition result in the convergence target variable value when detecting that the preset convergence condition is satisfied, further comprising:
acquiring the change rate of the illumination target component in the target variable in the iteration process, and judging whether the change rate is smaller than a preset threshold value;
if the change rate is smaller than a preset threshold value, judging that a preset convergence condition is met;
and if the change rate is not less than a preset threshold value, judging that the preset convergence condition is not met.
5. The intrinsic image decomposition method as claimed in claim 1, wherein the step of obtaining the preset constraint and the minimization objective function determined by the currently specified original image is preceded by the steps of:
receiving an image sequence decomposition instruction sent by a user, and acquiring the number of images and an image pixel matrix in an original image determined based on the image sequence decomposition instruction;
and determining the minimization objective function based on a preset constraint condition, the number of the images and an image pixel matrix.
6. The intrinsic image decomposition method as claimed in claims 1 to 5, further comprising:
rank punishment is carried out on the minimization target function based on the nuclear norm of the illumination target component in the minimization target function.
7. The method of intrinsic image decomposition according to claim 1, wherein said step of solving said minimized objective function based on an alternating direction multiplier and said preset constraint to obtain an intrinsic image decomposition result of said original image further comprises:
when the original image is a face image, taking the illumination intrinsic image in the intrinsic image decomposition result as a face characteristic image;
and carrying out face recognition on the face characteristic image based on a preset image recognition algorithm.
8. An intrinsic image decomposition device, comprising:
the target function determining module is used for acquiring a preset constraint condition and a minimized target function determined by the current specified original image, wherein the preset constraint condition is determined according to the low-rank sparse property of the illumination intrinsic image;
and the decomposition result acquisition module is used for solving the minimized objective function based on an alternating direction multiplier method and the preset constraint condition so as to acquire an intrinsic image decomposition result of the original image.
9. An intrinsic image decomposition apparatus characterized by comprising: memory, processor and an intrinsic image decomposition program stored on the memory and executable on the processor, which when executed by the processor implements the steps of the intrinsic image decomposition method according to any one of claims 1 to 7.
10. A computer readable storage medium, characterized in that the computer readable storage medium has stored thereon an intrinsic image decomposition program which, when executed by a processor, implements the steps of the intrinsic image decomposition method according to any one of claims 1 to 7.
CN201911256829.7A 2019-12-09 2019-12-09 Intrinsic image decomposition method, device, equipment and readable storage medium Pending CN110929686A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201911256829.7A CN110929686A (en) 2019-12-09 2019-12-09 Intrinsic image decomposition method, device, equipment and readable storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201911256829.7A CN110929686A (en) 2019-12-09 2019-12-09 Intrinsic image decomposition method, device, equipment and readable storage medium

Publications (1)

Publication Number Publication Date
CN110929686A true CN110929686A (en) 2020-03-27

Family

ID=69859407

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201911256829.7A Pending CN110929686A (en) 2019-12-09 2019-12-09 Intrinsic image decomposition method, device, equipment and readable storage medium

Country Status (1)

Country Link
CN (1) CN110929686A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112927200A (en) * 2021-02-23 2021-06-08 北京房江湖科技有限公司 Intrinsic image decomposition method and device, readable storage medium and electronic equipment

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104268897A (en) * 2014-10-28 2015-01-07 天津大学 Synergetic and intrinsic image decomposition method
CN104484884A (en) * 2014-12-30 2015-04-01 天津大学 Intrinsic image decomposition method based on multi-scale L0 sparse constraint
CN106296749A (en) * 2016-08-05 2017-01-04 天津大学 RGB D image eigen decomposition method based on L1 norm constraint

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104268897A (en) * 2014-10-28 2015-01-07 天津大学 Synergetic and intrinsic image decomposition method
CN104484884A (en) * 2014-12-30 2015-04-01 天津大学 Intrinsic image decomposition method based on multi-scale L0 sparse constraint
CN106296749A (en) * 2016-08-05 2017-01-04 天津大学 RGB D image eigen decomposition method based on L1 norm constraint

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
WENYONG GONG, WEIHONG XU: "Intrinsic Image Sequence Decomposition Using Low-Rank Sparse Model", 《IEEE ACCESS》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112927200A (en) * 2021-02-23 2021-06-08 北京房江湖科技有限公司 Intrinsic image decomposition method and device, readable storage medium and electronic equipment
CN112927200B (en) * 2021-02-23 2022-02-01 贝壳找房(北京)科技有限公司 Intrinsic image decomposition method and device, readable storage medium and electronic equipment

Similar Documents

Publication Publication Date Title
US11334763B2 (en) Image processing methods, training methods, apparatuses, devices, media, and programs
CN108460338B (en) Human body posture estimation method and apparatus, electronic device, storage medium, and program
EP3042361B1 (en) Estimating depth from a single image
US9904874B2 (en) Hardware-efficient deep convolutional neural networks
US9344690B2 (en) Image demosaicing
US10289927B2 (en) Image integration search based on human visual pathway model
CN112162930B (en) Control identification method, related device, equipment and storage medium
CN110555795A (en) High resolution style migration
CN108427927B (en) Object re-recognition method and apparatus, electronic device, program, and storage medium
CN109902763B (en) Method and device for generating feature map
CN109948700B (en) Method and device for generating feature map
CN109948699B (en) Method and device for generating feature map
US11604963B2 (en) Feedback adversarial learning
CN112489114A (en) Image conversion method and device, computer readable storage medium and electronic equipment
KR20200102409A (en) Key frame scheduling method and apparatus, electronic devices, programs and media
CN110263301B (en) Method and device for determining color of text
CN110532891B (en) Target object state identification method, device, medium and equipment
CN110929686A (en) Intrinsic image decomposition method, device, equipment and readable storage medium
Mantecón et al. New generation of human machine interfaces for controlling UAV through depth-based gesture recognition
CN113361384A (en) Face recognition model compression method, device, medium, and computer program product
CN109919249B (en) Method and device for generating feature map
CN117315758A (en) Facial expression detection method and device, electronic equipment and storage medium
CN113762173B (en) Training method and device for face optical flow estimation and optical flow value prediction model
CN112927200B (en) Intrinsic image decomposition method and device, readable storage medium and electronic equipment
CN110110111B (en) Method and device for monitoring screen

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
TA01 Transfer of patent application right

Effective date of registration: 20210825

Address after: 528000 room 314, block B, neifo high tech think tank center, Nanhai Software Science Park, Shishan town, Nanhai District, Foshan City, Guangdong Province

Applicant after: Foshan Shining Pupil Technology Co.,Ltd.

Address before: 528000 room 314, block B, neifo high tech think tank center, Nanhai Software Science Park, Shishan town, Nanhai District, Foshan City, Guangdong Province

Applicant before: Foshan Shining Pupil Technology Co.,Ltd.

Applicant before: Shenzhen Sancheng Zhichuang Technology Co.,Ltd.

TA01 Transfer of patent application right
RJ01 Rejection of invention patent application after publication

Application publication date: 20200327

RJ01 Rejection of invention patent application after publication