Error evaluation and prediction method and system for color constancy algorithm for indoor scene

Abstract

The application discloses an indoor scene-oriented color constancy algorithm error evaluation prediction method and system, and belongs to the technical field of computer and information service. The application comprises the following steps: acquiring an indoor scene color constancy dataset containing portrait content; acquiring real light source color information of a color cast image in the dataset; acquiring estimated light source color information obtained based on a color constancy algorithm to be evaluated; performing color space conversion on the light source color information; for the color constancy algorithm to be evaluated, corresponding estimated values are obtained according to the real light source color attribute, the estimated light source color attribute and a color constancy algorithm error evaluation quantitative model, so that the light source color estimation accuracy of different color constancy algorithms is represented. The application realizes high-precision and high-stability error evaluation of the color constancy algorithm for the indoor scene containing the portrait content.

Description

Technical Field

[0001] The present invention belongs to the field of computer and information service technology, and particularly relates to a color constancy algorithm error evaluation and prediction method and system for indoor scenes. Background Art

[0002] Color constancy refers to the perceptual property of the human visual system that allows the color of an object's surface to remain constant even when the color of the light illuminating it changes. However, computers lack the same visual properties as the human eye and cannot ignore the influence of light sources on the color of objects in an image. To enable computer vision systems to replicate this human-like property, the process of restoring the human eye's color constancy has become a major research topic in the computer field and is currently widely used in image processing, computer vision, and other fields.

[0003] Color constancy algorithms are a general term for methods that address computer color constancy. Their general principle is to estimate the color of the light source from the visual scene to facilitate image correction. Numerous color constancy algorithms exist, making error assessment of color constancy algorithms in specific scenarios particularly important. The basic principle of error assessment for color constancy algorithms is to calculate the degree of approximation between the original light source and the algorithm's estimated light source after processing a general dataset of images. Currently, a common method for quantifying this degree of approximation is to mathematically measure the difference between the original and estimated light sources in the normalized RGB color space. Commonly used mathematical metrics include restoration angle error, reproduction angle error, and Euclidean distance. Furthermore, because human vision is highly sensitive to the color of human faces, indoor scenes containing portraits often require special consideration and treatment in image processing and computer vision.

[0004] However, at present, relevant research by scholars at home and abroad has not fully considered the particularity of indoor scenes containing portrait content, and the error assessment prediction methods of existing color constancy algorithms have not overcome the interference of different light sources on the stability of error assessment, nor have they fully considered the impact of different color spaces on the accuracy of error assessment.

[0005] To address the above issues, a technical solution is urgently needed. For indoor scenes containing portraits, an appropriate color space should be used to implement an error assessment and prediction method and system for a color constancy algorithm that conforms to the subjective color visual perception of the human eye, while taking into account the stability of the error assessment results under different light sources. Summary of the Invention

[0006] The purpose of the present invention is to solve the problems described in the background art and to propose a method and system for evaluating and predicting color constancy algorithms for indoor scenes. For indoor scenes containing portraits, the human eye's visual attention is mainly focused on facial skin color and the surface color of indoor objects. Since the CIELAB color space is more accurate in representing surface color, the CIELAB color space should be used. However, since the white point of the CIELAB color space is (100, 0, 0) and contains zero values, and previous color constancy algorithm error evaluation methods all require the white point coordinates as the denominator for normalization, previous color constancy algorithm error evaluation methods cannot be calculated in the CIELAB color space, and therefore cannot achieve good prediction results for indoor scenes containing portraits. The technical solution proposed by the present invention can effectively solve the above-mentioned problems, thereby achieving high-precision and high-stability error evaluation of color constancy algorithms for indoor scenes containing portraits.

[0007] The technical solution of the present invention is to provide a color constancy algorithm error evaluation and prediction method for indoor scenes, comprising the following steps:

[0008] Step 1: Obtain a color constancy dataset of indoor scenes containing portrait content;

[0009] Step 2: Get the true light source color information RGB of the color-biased image in the dataset of step 1. u ;

[0010] Step 3: Obtain the estimated light source color information RGB calculated based on the color constancy algorithm to be evaluated e ;

[0011] Step 4: Convert the above light source color information to color space S1 to obtain the real light source color attribute C u and the estimated light source color attribute C calculated based on the color constancy algorithm to be evaluated e ;

[0012] Step 5: Convert the real light source color attribute C converted in step 4 u , the estimated light source color attribute C calculated based on the color constancy algorithm to be evaluated after conversion e Input into the color constancy algorithm error evaluation quantization model F constructed by the present invention to obtain the error estimation value of the color constancy algorithm to be evaluated, thereby realizing the characterization of the light source color estimation accuracy of different color constancy algorithms;

[0013] The specific form of the color constancy algorithm error evaluation quantization model F is as follows:

[0014]

[0015]

[0016] wherein F is the color constancy algorithm error estimation value, C w = (L w , a w , b w ) = (100, 0, 0) is the CIELAB color space reference white point, wherein the L value represents the lightness of the color, the a value represents the red-green component of the color, and the b value represents the yellow-blue component of the color, C u = (L u , a u , b u ) is the color attribute of the real light source color information converted to the color space S1, C e = (L e , a e , b e ) is the color attribute of the estimated light source color information converted to the color space S1 based on the calculation of the color constancy algorithm to be evaluated.

[0017] Moreover, in step 2, the real light source color information RGB u is adopted in the normalized RGB color space.

[0018] Moreover, in step 3, the color constancy algorithm to be evaluated includes Grey-Edge, GP, Grey-World, max-RGB, PCA, and Shades of Grey algorithms.

[0019] Moreover, in step 3, the estimated light source color information RGB e adopted in the normalized RGB color space based on the calculation of the color constancy algorithm to be evaluated.

[0020] Moreover, in step 4, the color space S1 adopts the CIELAB color space.

[0021] The present application also provides an error evaluation and prediction system for color constancy algorithms for indoor scenes, comprising the following modules:

[0022] A data set acquisition module for acquiring an indoor scene color constancy data set containing portrait content;

[0023] A real light source color acquisition module for acquiring real light source color information RGB u of a color cast image in the indoor scene color constancy data set containing portrait content;

[0024] An estimated light source color acquisition module for acquiring estimated light source color information RGB e based on the calculation of a color constancy algorithm to be evaluated.

[0025] a color space conversion module for converting the above light source color information to a color space S1 to obtain a real light source color attribute C u and an estimated light source color attribute C e calculated based on the color constancy algorithm to be evaluated.

[0026] a color constancy algorithm error evaluation module for evaluating the prediction error size of the color constancy algorithm to be evaluated using a color constancy algorithm error evaluation quantification model, and measuring the light source color estimation accuracy of the color constancy algorithm to be evaluated in the subjective color perception level according to the output result. The implementation is to input the converted real light source color attribute C u , the converted estimated light source color attribute C e calculated based on the color constancy algorithm to be evaluated into the color constancy algorithm error evaluation quantification model F constructed by the present application to obtain the error estimation value of the color constancy algorithm to be evaluated, and further realize the representation of the light source color estimation accuracy of different color constancy algorithms.

[0027] The specific form of the color constancy algorithm error evaluation quantification model F is as follows:

[0028]

[0029]

[0030] Wherein, F is the color constancy algorithm error estimation value, C w =(L w ,a w ,b w )=(100,0,0) is the CIELAB color space reference white point, wherein the L value represents the brightness of the color, the a value represents the red-green component of the color, and the b value represents the yellow-blue component of the color, C u =(L u ,a u ,b u ) is the color attribute of the real light source color information converted to the color space S1, C e =(L e ,a e ,b e ) is the color attribute of the estimated light source color information converted to the color space S1 based on the color constancy algorithm to be evaluated.

[0031] Moreover, in the real light source color collection module, the real light source color information RGB u adopts a normalized RGB color space.

[0032] Moreover, in the light source color estimation module, the color constancy algorithms to be evaluated include Grey-Edge, GP, Grey-World, max-RGB, PCA, and Shades of Grey algorithms.

[0033] Moreover, in the light source color acquisition module, the estimated light source color information RGB is calculated based on the color constancy algorithm to be evaluated. e Use normalized RGB color space.

[0034] Moreover, in the color space conversion module, the color space S1 adopts the CIELAB color space.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] The present invention proposes a color constancy algorithm error assessment and prediction technology solution for indoor scenes. Based on the characteristics of indoor scene images containing portrait content and using a color constancy algorithm error assessment quantification model, it achieves a comprehensive and accurate characterization of the light source color estimation effects of different color constancy algorithms, thereby providing an effective and targeted color constancy algorithm error assessment method for the field. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a flow chart of an embodiment of the present invention;

[0038] Figure 2 is an example diagram of the color constancy dataset used in the embodiment of the present invention. DETAILED DESCRIPTION

[0039] The following provides a detailed description of the embodiments of the present invention with reference to the accompanying drawings.

[0040] like Figure 1 As shown, an embodiment of the present invention provides a color constancy algorithm error assessment and prediction technology solution for indoor scenes. It relies on the characteristics of indoor scene images containing portrait content and uses a color constancy algorithm error assessment quantification model as a means to achieve a comprehensive and accurate characterization of the light source color estimation effects of different color constancy algorithms, thereby providing an effective and targeted color constancy algorithm error assessment method for the field.

[0041] The embodiment adopts the indoor scene part containing human images in the Gehler dataset in the color constancy general dataset, which contains 58 different images, each image contains a ColorChecker color card as a reference, and all pictures are saved in the Canon RAW format; and adopts six classic color constancy algorithms as the color constancy algorithm to be evaluated, including Grey-Edge, GP, Grey-World, max-RGB, PCA and Shades of Grey algorithms; the psychophysical experiment results are used as the model test basis to illustrate the accuracy of the error evaluation and prediction method of the color constancy algorithm proposed in the present application. It should be noted that the present application is not limited to the above color constancy dataset or color constancy algorithm, and the present method is also applicable to other datasets and algorithms.

[0042] The technical scheme of the present application can be automatically implemented by a person skilled in the art using computer software technology when implemented. The method process provided by the embodiment includes the following steps:

[0043] 1) Obtain an indoor scene color constancy dataset containing human image content;

[0044] In the embodiment, the indoor scene part containing human images in the Gehler dataset in the color constancy general dataset is adopted, which contains 58 different images, each image contains a ColorChecker color card as a reference, and all pictures are saved in the Canon RAW format. For example, as shown in the following table: Figure 2

[0045] 2) Obtain the true light source color information RGB u of the color cast image in the indoor scene color constancy dataset containing human image content, and adopt a normalized RGB color space;

[0046] In the embodiment, the true light source color information RGB u of the color cast image in the color constancy dataset in step 1) can be obtained by reading the RGB value corresponding to the white block in the ColorChecker color card in the image, and the value obtained after normalization is RGB u .

[0047] 3) Obtain the estimated light source color information RGB e calculated based on the color constancy algorithm to be evaluated, and adopt a normalized RGB color space;

[0048] ​In the embodiment, the color constancy algorithms to be evaluated include 6 classic color constancy algorithms as the color constancy algorithms to be evaluated, including Grey-Edge, GP, Grey-World, max-RGB, PCA, and Shades of Grey algorithms. The 6 algorithms are used to estimate the light source color for the color-cast image of the color constancy dataset selected in step 1), and the values ​​obtained after normalization are RGB values. e .

[0049] 4) Convert the above light source color information to the color space S1 to obtain the real light source color attribute C u and the estimated light source color attribute C calculated based on the color constancy algorithm to be evaluated e , color space S1 adopts CIELAB color space.

[0050] In the embodiment, the light source color information obtained in the above steps 2) and 3) is converted into the CIELAB color space to obtain the real light source color attribute C u and the estimated light source color attribute C calculated based on the color constancy algorithm to be evaluated e The above color space conversion method is a public method and will not be described in detail in the present invention.

[0051] 5) The real light source color attribute C converted in step 4) u , the estimated light source color attribute C calculated based on the color constancy algorithm to be evaluated after conversion e Input into the color constancy algorithm error evaluation quantization model F constructed by the present invention to obtain the error estimation value of the color constancy algorithm to be evaluated, thereby realizing the characterization of the light source color estimation accuracy of different color constancy algorithms;

[0052] The specific form of the color constancy algorithm error evaluation quantization model F is as follows:

[0053]

[0054]

[0055] Among them, F is the error estimation value of the color constancy algorithm, C w =(L w ,a w ,b w )=(100,0,0) is the reference white point of CIELAB color space, where L value represents the lightness of the color, a value represents the red-green component of the color, b value represents the yellow-blue component of the color, C u =(L u ,a u ,b u) is the color attribute of the real light source color information converted to the color space S1, C e =(L e ,a e ,b e ) is the color attribute converted into the color space S1 based on the estimated light source color information calculated based on the color constancy algorithm to be evaluated.

[0056] To further demonstrate the technical advantages of the method described in the present invention in error estimation of color constancy algorithms, a subjective comparison experiment was conducted. By calculating the weighted correlation coefficient Pearson's r of all color-cast images, the weighted correlation coefficient (Pearson's r) between the subjective evaluation values ​​of the degree of color cast after correction for the six color constancy algorithms obtained in the subjective experiment and the color constancy algorithm error assessment quantification model F in step 5) was calculated. The results were then compared with various color constancy algorithm error assessment methods previously used in this field. Specifically, the method was implemented as follows: using the 58 color-cast images of indoor scenes containing portraits from the aforementioned Gehler dataset as the materials, and using the six classic color constancy algorithms as the color constancy algorithms to be evaluated, a subjective color cast evaluation experiment was conducted on the color-cast images after correction for the algorithms. The specific experimental method is as follows:

[0057] 1) The monitor used in the experiment was a Sanc G7c 27-inch monitor with a resolution of 2560*1440, a 16:9 aspect ratio, a contrast ratio of 1000:1, a response time of 1ms, 129% sRGB color space coverage, and a refresh rate of 60Hz. To optimize image display, we used iColor Display3 software and an i1 Pro2 spectrophotometer for color management and calibration. This experiment used the sRGB standard preset, with a target white point temperature of 6500K and a gamma value of 2.2. The measured white point temperature was 6664K and a gamma value of 2.21, resulting in a white point temperature deviation of 2.52% and a gamma value deviation of 0.45%, both within the error range.

[0058] 2) A total of 12 identical observers participated in the experiment, including 6 male and 6 female observers. All observers were students and had passed the Ishihara color blindness test, indicating normal color vision. The observers ranged in age from 18 to 25 years old. After passing the test, they completed a questionnaire outside the laboratory containing information such as gender, age, and skin color. Upon entering the laboratory, the observers changed into a gray lab coat to prevent reflections and color interference from the clothes themselves, and were asked to turn off light-emitting devices such as mobile phones.

[0059] 3) The visual experiment judgment operation interface is mainly composed of an image area, a black background, a scoring area, and an image switching button, etc. The images in the interface are generated by converting the skew color images after generating a diagonal matrix from the estimated light source and the real light source in the data set, and then preprocessed. They respectively represent the image under the correct reference light source and the image under the reference light source obtained by the algorithm, and the correct image is distinguished by an inverted triangle symbol above it. The scoring area has 7 levels of scoring, and the observer selects the corresponding score before clicking the "Next Image" button to jump to the next pair of images. The experiment progress can be viewed in real time in the lower right corner.

[0060] 4) The experiment is carried out in a dark environment, and the display is the only light source. Before the experiment starts, the observer sits down under the guidance of the experimenter, adjusts the seat position and height so that he can look straight at the center of the display at a distance of 60 cm from the display. After the observer enters the color adaptation state, the experimenter explains the experiment in detail. The observer observes the two images on the display as required, compares their similarity, and selects the score from the following seven options: 1 point (very dissimilar), 2 points (relatively dissimilar), 3 points (a little dissimilar), 4 points (uncertain similarity or dissimilarity), 5 points (a little similar), 6 points (relatively similar), and 7 points (very similar). The observation time is not limited. After the observer determines the score, he clicks the "Next Image" button in the lower right corner to switch to the next group of images. During the experiment, the progress of the experiment will be displayed in the lower right corner, and a text reminder will be popped up after the experiment is completed, and the observer's information will be recorded. For each observer, the order of the images and the left-right order of the real image and the algorithm-processed image are random. A single experiment contains 6*58=348 evaluations and takes about 20 minutes.

[0061] The subjective experiment can obtain the subjective evaluation value of the observer for the color cast degree of the 58 indoor scene skew color images containing portraits corrected by the 6 color constancy algorithms to be evaluated, and further calculate the weighted correlation coefficient Pearson's r between the subjective evaluation value and the error evaluation quantitative model F of the color constancy algorithm constructed by the application, and compare it with the error evaluation methods of various color constancy algorithms in the prior art. The results show that the correlation coefficient between the subjective evaluation value and the estimated value of the model is 0.924, which is significantly higher than that of the prior art, proving that the color constancy algorithm error evaluation quantitative model F constructed by the application has high accuracy, and further proving that the method described in the application has strong technical advantages in the accuracy evaluation of color constancy algorithms.

[0062] Table 1 Weighted correlation coefficient Pearson's r test results

[0063]

[0064] The present invention also provides a color constancy algorithm error evaluation and prediction system for indoor scenes, comprising the following modules:

[0065] The dataset acquisition module is used to obtain a color constancy dataset of indoor scenes containing portrait content;

[0066] The real light source color acquisition module is used to obtain the real light source color information RGB of the color cast images in the indoor scene color constancy dataset containing portrait content u ;

[0067] The estimated light source color acquisition module is used to obtain the estimated light source color information RGB calculated based on the color constancy algorithm to be evaluated e ;

[0068] The color space conversion module is used to convert the above light source color information into the color space S1 to obtain the real light source color attribute C u and the estimated light source color attribute C calculated based on the color constancy algorithm to be evaluated e ;

[0069] The color constancy algorithm error evaluation module is used to evaluate the prediction error of the color constancy algorithm to be evaluated using the color constancy algorithm error evaluation quantification model, and to measure the light source color estimation accuracy of the color constancy algorithm to be evaluated at the subjective color perception level based on the output results. The implementation method is to convert the converted real light source color attribute C u , the estimated light source color attribute C calculated based on the color constancy algorithm to be evaluated after conversion e The error is input into the color constancy algorithm error evaluation quantization model F constructed by the present invention to obtain the error estimation value of the color constancy algorithm to be evaluated, thereby realizing the characterization of the light source color estimation accuracy of different color constancy algorithms.

[0070] The specific form of the color constancy algorithm error evaluation quantization model F is as follows:

[0071]

[0072]

[0073] Among them, F is the error estimation value of the color constancy algorithm, C w =(L w ,a w ,b w)=(100,0,0) is the reference white point of CIELAB color space, where L value represents the lightness of the color, a value represents the red-green component of the color, b value represents the yellow-blue component of the color, C u =(L u ,a u ,b u ) is the color attribute of the real light source color information converted to the color space S1, C e =(L e ,a e ,b e ) is the color attribute converted into the color space S1 based on the estimated light source color information calculated based on the color constancy algorithm to be evaluated.

[0074] The specific implementation of each module corresponds to each step, which will not be described in detail in the present invention.

[0075] The specific embodiments described herein are merely illustrative of the spirit of the present invention. Persons skilled in the art may make various modifications, additions, or substitutions to the described specific embodiments without departing from the spirit of the present invention or exceeding the scope of the appended claims.

Claims

1. A color constancy algorithm error evaluation and prediction method for indoor scenes, characterized by: The following steps are involved: Step 1: Obtain a color constancy dataset of indoor scenes containing portrait content; Step 2: Get the true light source color information RGB of the color-biased image in the dataset of step 1. u ; Step 3: Obtain the estimated light source color information RGB calculated based on the color constancy algorithm to be evaluated e ; Step 4: Convert the above light source color information to color space S1 to obtain the real light source color attribute C u and the estimated light source color attribute C calculated based on the color constancy algorithm to be evaluated e ; Step 5: Set the real light source color attribute C u , estimate the light source color attribute C e Input it into the constructed color constancy algorithm error evaluation quantization model F to obtain the error estimation value of the color constancy algorithm to be evaluated; The specific form of the color constancy algorithm error evaluation quantization model F is as follows: Among them, F is the error estimation value of the color constancy algorithm, C w =(L w ,a w ,b w )=(100,0,0) is the reference white point of CIELAB color space, where L value represents the lightness of the color, a value represents the red-green component of the color, b value represents the yellow-blue component of the color, C u =(L u ,a u ,b u ) is the color attribute of the real light source color information converted to the color space S1, C e =(L e ,a e ,b e ) is the color attribute converted into the color space S1 based on the estimated light source color information calculated based on the color constancy algorithm to be evaluated.

2. The color constancy algorithm error evaluation and prediction method for indoor scenes according to claim 1, characterized in that: In step 2, the real light source color information RGB u Use normalized RGB color space.

3. The color constancy algorithm error evaluation and prediction method for indoor scenes according to claim 1, characterized in that: In step 3, the color constancy algorithms to be evaluated include Grey-Edge, GP, Grey-World, max-RGB, PCA, and Shades of Grey algorithms.

4. The method for evaluating and predicting color constancy algorithm errors for indoor scenes according to claim 1, characterized in that: In step 3, the estimated light source color information RGB is calculated based on the color constancy algorithm to be evaluated e Use normalized RGB color space.

5. The color constancy algorithm error evaluation and prediction method for indoor scenes according to claim 1, characterized in that: In step 4, the color space S1 adopts the CIELAB color space.

6. A color constancy algorithm error evaluation and prediction system for indoor scenes, characterized by: Includes the following modules: The dataset acquisition module is used to obtain a color constancy dataset of indoor scenes containing portrait content; The real light source color acquisition module is used to obtain the real light source color information RGB of the color-biased image in the dataset u ; The estimated light source color acquisition module is used to obtain the estimated light source color information RGB calculated based on the color constancy algorithm to be evaluated e ; The color space conversion module is used to convert the above light source color information into the color space S1 to obtain the real light source color attribute C u and the estimated light source color attribute C calculated based on the color constancy algorithm to be evaluated e ; The color constancy algorithm error evaluation module is used to convert the real light source color attribute C u , estimate the light source color attribute C e Input it into the constructed color constancy algorithm error evaluation quantization model F to obtain the error estimation value of the color constancy algorithm to be evaluated; The specific form of the color constancy algorithm error evaluation quantization model F is as follows: Among them, F is the error estimation value of the color constancy algorithm, C w =(L w ,a w ,b w )=(100,0,0) is the reference white point of CIELAB color space, where L value represents the lightness of the color, a value represents the red-green component of the color, b value represents the yellow-blue component of the color, C u =(L u ,a u ,b u ) is the color attribute of the real light source color information converted to the color space S1, C e =(L e ,a e ,b e ) is the color attribute converted into the color space S1 based on the estimated light source color information calculated based on the color constancy algorithm to be evaluated.

7. The color constancy algorithm error evaluation and prediction system for indoor scenes according to claim 6, characterized in that: In the real light source color acquisition module, the real light source color information RGB u Use normalized RGB color space.

8. The color constancy algorithm error evaluation and prediction system for indoor scenes according to claim 6, characterized in that: In the light source color estimation module, the color constancy algorithms to be evaluated include Grey-Edge, GP, Grey-World, max-RGB, PCA, and Shades of Grey algorithms.

9. The color constancy algorithm error evaluation and prediction system for indoor scenes according to claim 6, characterized in that: In the light source color acquisition module, the estimated light source color information RGB is calculated based on the color constancy algorithm to be evaluated e Use normalized RGB color space.

10. The color constancy algorithm error evaluation and prediction system for indoor scenes according to claim 6, characterized in that: In the color space conversion module, the color space S1 adopts the CIELAB color space.

Images