CN116862895A - CCD quality evaluation method based on SSIM algorithm - Google Patents

Abstract

The invention discloses a CCD quality evaluation method based on an SSIM algorithm, which comprises the following steps: s1, acquiring an original image of a CCD by using a CCD quality evaluation device, and performing compression blurring processing on image data to obtain a blurred image; s2, generating a defect-free reconstruction image of the fuzzy image through a super-resolution reconstruction method; and S3, comparing the original image with the reconstructed image based on an SSIM algorithm, and evaluating the quality of the CCD. The invention solves the defects of incomplete evaluation test process and non-uniform evaluation standard in the existing evaluation method.

Description

CCD quality evaluation method based on SSIM algorithm

Technical Field

The invention belongs to the technical field of surface defect detection, and particularly relates to a CCD quality evaluation method based on an SSIM algorithm.

Background

The infrared thermal imager needs to be assisted by using the blackbody in the process of performing the external field test debugging, and the traditional blackbody and related calibration equipment are large in volume and mass, so that the number of related equipment is large, the external field maneuvering deployment is inconvenient, and the infrared thermal imager is unfavorable for being unfolded and used in the wild at any time. Nowadays, infrared devices such as small infrared thermometers based on Charge Coupled Devices (CCDs) are beginning to be put into production and living on a large scale. The miniaturized or handheld infrared devices do not need to use traditional large blackbody, collimator and other devices when calibrating, correcting and evaluating. However, with the increase of the service life, the CCD is inevitably defective, aged, and the like, so that the quality of the CCD needs to be tested, and the quality of the CCD is generally evaluated by evaluating the quality of the infrared image photographed by the CCD.

The accurate evaluation of the infrared image quality is a vital work in performing the performance evaluation of the infrared system and guiding the improvement direction of the infrared system. The infrared image quality evaluation is divided into a subjective evaluation method and an objective evaluation method, the common subjective evaluation methods comprise an observer average score method, a viewpoint average score method and the like, a plurality of repeated tests are generally required to be carried out on a test image by a plurality of observers, the measurement process is tedious and time-consuming, the test result is greatly influenced by subjective factors such as knowledge background and experience of the tester, the stability is poor, and the image quality is difficult to evaluate rapidly and accurately, so that the application occasions are greatly limited. Therefore, an objective quality evaluation method for measuring image quality according to a quantization model has become an important research point in the field of image quality evaluation.

The image quality objective evaluation standard comprises indexes such as resolution, signal to noise ratio and the like, and conventional calculation methods of the indexes all need to rely on the known real image of the detection object, namely the image quality evaluation with reference. In practical applications, it is often very difficult or even impossible to obtain the reference image information. Therefore, there is an urgent need for a non-reference evaluation method, i.e., independent of reference image information, to calculate the characteristics of the image to be measured to obtain the image quality related information, thereby completing the image quality evaluation. There has been some research effort in this regard, and various research teams have proposed various characteristic indices that characterize image quality from different angles. Although the existing method can reflect the quality of the infrared image to a certain extent, the existing method still has great problems:

the first research team and the second research team put forward various evaluation criteria, which are incompatible with each other, especially with conventional performance indexes such as resolution, signal to noise ratio and other undetermined association relations, and the problem that the physical performance of the infrared imaging system is unable to be reflected directly from the evaluation indexes.

The second, the evaluation method is related to image characteristics, when evaluating a plurality of images of the same target, the quality of the image can be accurately obtained, but when evaluating images of different targets in different scenes, the images are difficult to directly compare with each other, so that unified evaluation standards applicable to different targets of different systems cannot be achieved.

Disclosure of Invention

The embodiment of the invention aims to provide a CCD quality evaluation method based on an SSIM algorithm, which aims to solve the defects of incomplete evaluation test process and non-uniform evaluation standard in the existing evaluation method.

In order to solve the technical problems, the technical scheme adopted by the invention is that the CCD quality evaluation method based on the SSIM algorithm comprises the following steps:

s1, acquiring an original image of a CCD by using a CCD quality evaluation device, and performing compression blurring processing on image data to obtain a blurred image;

s2, generating a defect-free reconstruction image of the fuzzy image through a super-resolution reconstruction method;

and S3, comparing the original image with the reconstructed image based on an SSIM algorithm, and evaluating the quality of the CCD.

Further, the CCD quality evaluation device in the step S1 comprises a standard blackbody target, a portable collimator is arranged between the standard blackbody target and a thermal imager, and the thermal imager is connected with a PC through an image adapter plate;

the standard blackbody target structure includes: the device comprises a peltier, a copper plate and a blackbody, wherein the peltier, the copper plate and the blackbody are sequentially overlapped, and the peltier is used for temperature control; the copper plate is used for uniformly distributing the temperature; the blackbody is used for imaging in front of the measured object and provides a target for the thermal imager.

Further, the compression blurring process in the step S1 is expressed as:

g(x，y)＝f(x，y)*h(x，y)+n(x，y)

where f (x, y) is the original image, h (x, y) is the blurred point spread function, n (x, y) is the detection noise, and g (x, y) represents the blurred image.

Further, the step S2 specifically includes:

s21, weakening the defect problem of blind pixels and non-uniformity in the blurred image;

s22, obtaining a defect-free reconstructed image of the estimated body by using a super-resolution reconstruction method;

and S23, evaluating the reconstructed image, and selecting the reconstructed image with the highest definition to input into the step S3.

Further, in the super-resolution reconstruction method in step S22, a two-phase kernel estimation algorithm is used for performing a fast kernel initialization, and the fast kernel initialization is used as a high-efficiency robust kernel estimation process; then restoring the original image with clear edge information by utilizing the space sparse characteristic; in the nuclear refining stage, an iterative support detection algorithm is adopted to check sparse limiting conditions and accurately reserve large-value elements; finally, an objective function of TV-l1 is adopted to be robust to noise, and an efficient operator based on half-quadratic division is used to reconstruct an initial clear reconstructed image.

Further, the evaluation method in the step S23 includes spatial domain feature evaluation, frequency domain feature evaluation, entropy function evaluation, and evaluation by constructing a composite index by spatial domain feature evaluation, frequency domain feature evaluation, entropy function evaluation.

Further, the airspace feature evaluation refers to extracting edge information of an image by using a gradient function, wherein the edge of the image is sharper when the gradient value is larger, and the visual sense is clearer; wherein the gradient function includes a Brenner function, a gradient square function, and a Variance function.

Further, the frequency domain feature evaluation means that after the image is subjected to fourier transformation, the more high-frequency components contained in the image, the clearer the corresponding image.

Further, the definition of the entropy function is:

wherein the method comprises the steps of，P _k Indicating the probability of occurrence of the kth region ambiguity.

Further, the specific expression of the SSIM algorithm in step S3 is:

SSIM(x,y)＝l(x,y)×c(x,y)×s(x,y)

wherein, l (x, y) is the brightness contrast function of the reconstructed image and the original image, c (x, y) is the contrast function of the reconstructed image and the original image, s (x, y) is the structure contrast function of the reconstructed image and the original image;

the larger the SSIM value range [0,1], the smaller the image distortion, indicating a higher CCD quality.

The beneficial effects of the invention are as follows:

the invention realizes the data generation, comparison and evaluation of various CCD sheet surface defects based on the SSIM algorithm and the standard blackbody device, improves the defect that the existing CCD quality evaluation device lacks uniform dimension, has better universal applicability, and simultaneously solves the defects of incomplete evaluation test process and non-uniform evaluation standard in the existing method.

Drawings

In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a schematic diagram of the system composition of a CCD quality evaluation device based on the SSIM algorithm according to an embodiment of the present invention.

Fig. 2 is a flow chart of reconstructed image evaluation according to an embodiment of the present invention.

Fig. 3 is a schematic structural diagram of an SSIM algorithm according to the present invention.

Detailed Description

The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The embodiment of the invention provides a CCD quality evaluation device based on an SSIM algorithm, which comprises a standard blackbody target, a thermal imager, a portable collimator and a PC, wherein the standard blackbody target has the following structure:

the temperature control device is formed by sequentially superposing a peltier, a copper plate and a blackbody part with temperature characteristics, wherein the peltier is used for temperature control; the copper plate is used for uniformly distributing the temperature; the blackbody is used for imaging in front of the measured object and provides a target for the thermal imager.

The standard blackbody target is driven by electric power, the temperature is controlled by the singlechip, the temperature is stabilized at 25 ℃, and the test lens provides the standard irradiation target.

The algorithm adopted in the embodiment operates in a Field Programmable Gate Array (FPGA) in a PC, an input image is acquired by a camera link interface, referring to fig. 1, a data receiving and scheduling module receives line image data issued from a data transmission control module, and the line image data is temporarily stored in a DDR data buffer area, and the DDR data buffer area adopts a FIFO first-in first-out mode; meanwhile, the evaluation index serial communication mode is output, and the dispatching function of camera output data is realized.

The algorithm adopted by the embodiment is a structural similarity SSIM algorithm, and after the input image is subjected to fuzzy reconstruction, the structural similarity is used as an output index to indicate CCD quality evaluation parameters; meanwhile, the evaluation index serial communication mode is output, and the dispatching function of camera output data is realized.

The standard blackbody target, the thermal imager, the portable collimator and the PC are integrated, so that the function of assisting in debugging the thermal imager by using the blackbody system under the condition of an outfield test is realized. Meanwhile, the integrated blackbody system is also convenient for the relevant calibration and evaluation of the miniaturized and handheld thermal imager. In addition, because of the integration and portability of the system, the thermal imaging equipment can be calibrated, detected and evaluated in real time in related occasions.

The embodiment of the invention also provides a CCD quality evaluation method based on the SSIM algorithm, which comprises the steps of collecting images received by the CCD, compressing and blurring image data, generating a defect-free image through a super-resolution reconstruction technology, comparing the images based on the SSIM algorithm, and evaluating the quality of possible defects and ageing in the CCD. The method specifically comprises the following steps:

step 1: shooting a standard blackbody to form a unified noise-free shooting target, acquiring image data by a camera, storing the image data in a Random Access Memory (RAM) of the device, then entering an evaluation algorithm, and enabling a CCD quality evaluation device based on the SSIM algorithm to comprise a standard blackbody target and an evaluation device; the specific process is as follows:

the CCD equipment to be tested is arranged in a quality evaluation device, a uniform black body with the standard temperature of 25 ℃ is placed in front of the CCD equipment to be tested, the surface of the black body is required to have higher uniformity, and the temperature is monitored in real time in the evaluation process so as to unify dimensions.

(1-2) when the infrared target passes through the imaging system, the infrared target is limited by diffraction limit and the like, and the finally obtained image is a result of the original target image being blurred to a certain degree, and the process can be described by the following formula:

g(x，y)＝f(x，y)*h(x，y)+n(x，y)

where f (x, y) is the original image, h (x, y) is the blurred point spread function (Point Spread Function, PSF), n (x, y) is the detection noise, and g (x, y) represents the blurred image. The blurred image may in fact be represented as a convolution of the original image with the blurred point spread function in the spatial domain, which is also a linear transformation in nature.

According to the compressed sensing theory, when the point spread function h is known, solving the linear equation under the sparse constraint, namely inverting the clear image f through the blurred image g, wherein the process belongs to super-resolution reconstruction. The point spread function is unknown in the image resolution evaluation, but for a typical imaging system the nature of the system point spread function is predictable, with the specific function being determined by the imaging resolution. Therefore, corresponding point spread functions can be obtained under different resolution assumptions, and image super-resolution reconstruction is carried out according to the point spread functions, so that a series of reconstructed images are obtained. The clearest imaging results will be obtained when the point spread function used is consistent with the actual transfer function of the system.

(1-3) weakening the defects such as blind pixels and non-uniformity in the blurred image by using a point spread function and Fourier transformation.

(1-4) obtaining an image of the object to be evaluated, which should have a non-defect, by using a super-resolution reconstruction method, wherein a TV deconvolution model can be adopted in the super-resolution reconstruction process, so that a fuzzy kernel (blur kernel) can be well estimated, and non-blind convolution can be realized.

Firstly, carrying out quick kernel initialization through a two-phase kernel estimation algorithm to serve as a high-efficiency robust kernel estimation process; then restoring the original image with clear edge information by utilizing the space sparse characteristic; in the nuclear refining stage, an iterative support detection (Iterative Support Detection, ISD) algorithm is adopted to adaptively check sparse limiting conditions and accurately reserve large-value elements, and the idea of a soft threshold value is also used in the step; finally, the objective function of TV-l1 is adopted to be robust to noise, and an efficient operator based on half-quadratic division is used to reconstruct an initial clear image.

The resolution performance of the image can be obtained through evaluation of the super-resolution reconstruction result. In addition, the method obtains a resolution index which is irrelevant to specific image characteristics, the evaluation results of different imaging systems and different scene images can be compared with each other, and the evaluation process is shown in fig. 2. In order to judge whether the super-resolution reconstruction meets the requirement, the definition evaluation of the super-resolution reconstructed image is completed through a plurality of statistics based on image characteristics, and the adopted evaluation method comprises the following steps:

1-4-1 spatial signature: the theoretical basis of the brunner function, the gradient square function, the Variance function and the like is that a clear image contains sharper edge information, the edge information of the image is extracted by utilizing various gradient functions, and the sharper the gradient value is, the sharper the edge of the image is visually considered to be clear. Taking the example of a gradient square function, it is defined as the sum of the gradient values (sum of squares of gray differences with upper, lower, left and right pixels) of all pixels:

in the formula, I (x+1, y) -I (x, y) is a horizontal gray level difference, I (x, y+1) -I (x, y) is a vertical gray level difference, so that the square sum is a gray level square sum, and x and y respectively represent the horizontal coordinate and the vertical coordinate of the pixel.

1-4-2 frequency domain features: the clear image contains more detail information, and after the image is subjected to Fourier transform, the detail information of the image corresponding to the high-frequency component, namely the image corresponding to the more high-frequency component is clearer. Common spectral functions are high frequency component functions, threshold integral functions, wavelet functions, etc.

1-4-3 entropy function. The gray scale diversity of the blurred image is less than that of the clear image, and the characteristic is that the information entropy of the blurred image and the information entropy of the blurred image are different, so that the definition of the image can be evaluated by using the size of an entropy function of the image. The definition of the entropy function is:

P _k indicating the probability of occurrence of the kth region ambiguity.

In the actual evaluation of the sharpness of the image, the above method may be used to construct a composite index. The construction modes of the composite index include factor analysis, principal component analysis, regression analysis, cluster analysis, genetic algorithm and the like. The research shows that the performance of the composite index formed by combining a plurality of image quality evaluation indexes is superior to that of a single index by using a multivariate statistical analysis theory. For example, the above obtained portions are weighted after normalization, such as q=a×f (I) +b×e+c×f, where E is a frequency domain feature, and a, b, and c are weight ratios of spatial domain feature, frequency domain feature, and entropy function evaluation method, respectively, to finally obtain a composite index Q of the sharpness of the reconstructed image.

When the compressed sensing super-resolution reconstruction is utilized to obtain the resolution index, the precondition is that a clear image can be obtained through sparse solution. However, the super-resolution capability of any method is limited, and when the image resolution is low, the super-resolution to be achieved is too high, and such a resolution evaluation method may not be applicable. Thus, statistical methods such as spatial domain, frequency domain, entropy functions, etc., as described above, can be employed for lower resolution image evaluation. In fact, when the image resolution is low, the statistical evaluation methods are less affected by the image characteristics, and the evaluation result of the image resolution can be accurately given. When the image resolution is higher, the result given by the statistical method is related to the image characteristics, and at the moment, a more uniform resolution evaluation standard can be given by using the compressed sensing super-resolution reconstruction method. Therefore, the two methods are complementary in advantages, and can give more accurate and uniform resolution performance indexes at different image resolution levels.

After obtaining the optimal reconstruction resolution image, the structural consistency index of the reconstruction image and the original image can be used for comparison, and the parameters which evaluate the CCD quality and are irrelevant to the image, namely the signal-to-noise ratio difference between the original image and the super resolution reconstruction image, are obtained.

Structural consistency SSIM. Structural similarity (Structural similarity, SSIM) is a fully referenced image quality assessment indicator that measures image similarity from three aspects of brightness, contrast, and structure, respectively, see fig. 3. The larger the SSIM value range [0,1], the smaller the image distortion is, which shows that the higher the CCD quality is, the specific expression is:

SSIM(x,y)＝l(x,y)×c(x,y)×s(x,y)

wherein, l (x, y) is the brightness contrast function of the reconstructed image and the original image, c (x, y) is the contrast function of the reconstructed image and the original image, and s (x, y) is the structure contrast function of the reconstructed image and the original image.

From the definition above, the general image signal-to-noise ratio evaluation index needs to obtain the target original image, but in the no-reference evaluation, it is obvious that the target original image cannot be accurately obtained, but the infrared image a 'can be filtered and denoised through super-resolution reconstruction, the filtered image a "is approximately regarded as the original noise-free image, the approximate noise images are calculated from the images a" and a', and then an approximate signal-to-noise ratio is calculated through the images, the signal-to-noise ratio is not strictly equal to the actual signal-to-noise ratio, but the signal-to-noise ratio is reflected, the influence of noise on the signal can be expressed to a certain extent, and the actual signal-to-noise ratio is reflected.

In this specification, each embodiment is described in a related manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, with reference to the description of method embodiments in part.

The foregoing description is only of the preferred embodiments of the present invention and is not intended to limit the scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims (10)

1. The CCD quality evaluation method based on the SSIM algorithm is characterized by comprising the following steps of:

s1, acquiring an original image of a CCD by using a CCD quality evaluation device, and performing compression blurring processing on image data to obtain a blurred image;

s2, generating a defect-free reconstruction image of the fuzzy image through a super-resolution reconstruction method;

and S3, comparing the original image with the reconstructed image based on an SSIM algorithm, and evaluating the quality of the CCD.

2. The method for evaluating the quality of the CCD based on the SSIM algorithm according to claim 1, wherein the CCD quality evaluating device in the step S1 comprises a standard blackbody target, a portable collimator is arranged between the standard blackbody target and a thermal imager, and the thermal imager is connected with a PC through an image adapter plate;

the standard blackbody target structure includes: the device comprises a peltier, a copper plate and a blackbody, wherein the peltier, the copper plate and the blackbody are sequentially overlapped, and the peltier is used for temperature control; the copper plate is used for uniformly distributing the temperature; the blackbody is used for imaging in front of the measured object and provides a target for the thermal imager.

3. The method for evaluating the quality of a CCD based on the SSIM algorithm according to claim 1, wherein the compression blur process in step S1 is expressed as:

g(x，y)＝f(x，y)*h(x，y)+n(x，y)

where f (x, y) is the original image, h (x, y) is the blurred point spread function, n (x, y) is the detection noise, and g (x, y) represents the blurred image.

4. The method for evaluating the quality of the CCD based on the SSIM algorithm according to claim 1, wherein the step S2 is specifically:

s21, weakening the defect problem of blind pixels and non-uniformity in the blurred image;

s22, obtaining a defect-free reconstructed image of the estimated body by using a super-resolution reconstruction method;

and S23, evaluating the reconstructed image, and selecting the reconstructed image with the highest definition to input into the step S3.

5. The method for evaluating the quality of a CCD based on an SSIM algorithm according to claim 4, wherein in the super-resolution reconstruction method in step S22, a two-phase kernel estimation algorithm is used for performing a fast kernel initialization as a high-efficiency robust kernel estimation process; then restoring the original image with clear edge information by utilizing the space sparse characteristic; in the nuclear refining stage, an iterative support detection algorithm is adopted to check sparse limiting conditions and accurately reserve large-value elements; finally, an objective function of TV-l1 is adopted to be robust to noise, and an efficient operator based on half-quadratic division is used to reconstruct an initial clear reconstructed image.

6. The method according to claim 4, wherein the evaluation method in step S23 includes spatial domain feature evaluation, frequency domain feature evaluation, entropy function evaluation, and the evaluation is performed by constructing a composite index by spatial domain feature evaluation, frequency domain feature evaluation, entropy function evaluation.

7. The method for evaluating the quality of the CCD based on the SSIM algorithm according to claim 6, wherein the spatial domain feature evaluation is to extract the edge information of the image by using a gradient function, and the edge of the image is sharper and more visually clear when the gradient value is larger; wherein the gradient function includes a Brenner function, a gradient square function, and a Variance function.

8. The method for evaluating the quality of a CCD based on an SSIM algorithm according to claim 6, wherein the frequency domain feature evaluation means that the more high-frequency components contained in the image, the clearer the corresponding image is after the image is subjected to Fourier transform.

9. The method for evaluating the quality of a CCD based on an SSIM algorithm according to claim 6, wherein the entropy function is defined as:

wherein P is _k Indicating the probability of occurrence of the kth region ambiguity.

10. The method for evaluating the quality of the CCD based on the SSIM algorithm according to claim 1, wherein the specific expression of the SSIM algorithm in the step S3 is as follows:

SSIM(x,y)＝l(x,y)×c(x,y)×s(x,y)

wherein, l (x, y) is the brightness contrast function of the reconstructed image and the original image, c (x, y) is the contrast function of the reconstructed image and the original image, s (x, y) is the structure contrast function of the reconstructed image and the original image;

the larger the SSIM value range [0,1], the smaller the image distortion, indicating a higher CCD quality.