A low-light image enhancement quality evaluation method based on dark channel

Abstract

The present invention discloses a low-light image enhancement quality evaluation method based on dark channel, and the steps are as follows: using a dark channel extraction model to obtain the dark channel of a low-light enhanced image and a reference image; using a CNN neural network to extract features of illumination information and structural information to obtain dark channel depth features; using a CNN neural network to extract features of the low-light enhanced image and the reference image to obtain the overall features of the image; comparing the similarity of the dark channel and overall features of the low-light enhanced image and the reference image; and obtaining the final quality score through training and learning. The present invention is aimed at low-light image enhancement quality evaluation, and uses a dark channel method to extract the dark channel information of the image. Compared with general image quality evaluation methods, the present invention takes into account the image illumination information, and can achieve better evaluation effects and evaluation performance, and can also assist in improving the processing effect of existing low-light image enhancement algorithms.

Description

Technical Field

[0001] The present invention relates to the technical field of image quality evaluation and image enhancement, in particular to the technical field of low-light image quality evaluation, and specifically to a low-light image enhancement quality evaluation method based on a dark channel. Background Art

[0002] Low-light images often result in unclear visual effects and blurred image details. Low-light image enhancement is an important research direction in the field of image processing. It hopes to use algorithms to enhance underexposed images to a range that is tolerable to the human eye and the algorithm, thereby mining valuable information from these images and improving the performance of computer vision systems. Low-light image enhancement technology has many application scenarios and plays an indispensable role in computer vision, machine learning, security detection, medical imaging, space exploration, deep-sea exploration and other fields. Low-light images have diverse content, complex low-light scenes, and additional interference during image acquisition. Therefore, existing low-light image enhancement algorithms may have problems such as over / underexposure, inaccurate color restoration, and loss of details. A good evaluation standard is needed to help and guide low-light image enhancement.

[0003] At the same time, although there are many indicators for image quality evaluation, there is a lack of corresponding methods for the problem of low-light image enhancement quality evaluation. Summary of the invention

[0004] The purpose of the present invention is to solve the above-mentioned defects in the prior art and provide a low-light image enhancement quality evaluation method based on a dark channel.

[0005] The purpose of the present invention can be achieved by adopting the following technical solutions:

[0006] A low-light image enhancement quality evaluation method based on a dark channel, the low-light image enhancement quality evaluation method comprising the following steps:

[0007] Input a low-light enhanced image and a reference image, and obtain the dark channels of the low-light enhanced image and the reference image through a dark channel extraction model;

[0008] Using a CNN neural network to extract features of the dark channels of the low-light enhanced image and the reference image, and obtaining dark channel depth features of the low-light enhanced image and the reference image respectively;

[0009] Using a CNN neural network to extract features from the low-light enhanced image and the reference image, respectively obtaining overall features of the low-light enhanced image and the reference image;

[0010] comparing the similarity of dark channel depth features and overall features of the low-light enhanced image and the reference image;

[0011] According to the similarity of the dark channel depth features and overall features of the low-light enhanced image and the reference image, the final quality score is obtained through training and learning.

[0012] Furthermore, the dark channel extraction model expression is as follows:

[0013]

[0014] Among them, J dark () represents dark channel information, J C () represents the haze-free image of channel C, C represents one of the three channels R, G, and B, and x represents the input image.

[0015] Furthermore, the CNN neural network performs feature extraction on the dark channels of the low-light enhanced image and the reference image to extract dark channel features representing illumination information. The feature extraction process is as follows:

[0016] The deep features corresponding to the dark channels of the low-light enhanced image and the reference image are calculated respectively through the VGGNetCNN neural network (Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition [J]. arXiv, 2014: 1409.1556.);

[0017] In the VGGNetCNN neural network, the outputs of the five hidden convolutional layers are taken as the depth features corresponding to the dark channels of the low-light enhanced image and the reference image;

[0018] Through the VGGNetCNN neural network, 5 dark channel depth features of the low-light enhanced image and the reference image are obtained respectively.

[0019] Furthermore, the CNN neural network performs feature extraction on the low-light enhanced image and the reference image to extract general overall image features. The feature extraction process is as follows:

[0020] The deep features corresponding to the low-light enhanced image and the reference image are calculated through the VGGNetCNN neural network;

[0021] In the VGGNetCNN neural network, the outputs of the five hidden convolutional layers are taken as the overall features of the low-light enhanced image and the reference image respectively;

[0022] Through the VGGNetCNN neural network, 5 overall features of the low-light enhanced image and the reference image are obtained.

[0023] Furthermore, the similarity between the dark channel depth features and the overall features of the low-light enhanced image and the reference image is calculated to measure the similarity between the images for quality evaluation. The calculation formula for the similarity between the dark channel depth features and the overall features of the low-light enhanced image and the reference image is as follows:

[0024]

[0025]

[0026] Where i represents the i-th feature, i = 1, 2, 3, 4, 5, and Respectively represent the similarity between the i-th dark channel depth feature and the i-th overall feature, and They represent the i-th dark channel depth feature and the i-th overall feature of the low-light enhanced image, respectively. and They represent the i-th dark channel depth feature and the i-th overall feature of the reference image respectively.

[0027] Furthermore, based on the similarity between the dark channel depth feature and the overall feature, the quality score is calculated to evaluate the effect of the low-light enhancement algorithm. The final quality score obtained through training and learning is: S = w dark D dark +w I D I

[0028] Among them, w dark and w I are the scoring weights of the dark channel depth feature similarity and the overall feature similarity obtained through training, D dark and D I They are the overall similarity of the dark channel depth features and the overall similarity of the overall features, namely:

[0029] Compared with the prior art, the present invention has the following advantages and effects:

[0030] (1) The present invention uses dark channel information to obtain the illumination information of the image, and uses the dark channel information and the illumination information to evaluate the quality of the low-light enhanced image, which can produce better low-light enhanced image quality evaluation results.

[0031] (2) Dark channel information can be used as an evaluation index for image blur and interference. It can be used for low-light image enhancement quality evaluation to better determine whether blur and interference still exist after enhancement. It can also better evaluate the quality of low-light enhanced images and better assist in improving low-light image enhancement algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0033] Figure 1 The present invention discloses a low-light image enhancement quality evaluation method based on a dark channel;

[0034] Figure 2 is a schematic diagram of dark channel information of an input low-light enhanced image and a reference image in an embodiment of the present invention; wherein,

[0035] Figure 2 (a) is a schematic diagram of low-light enhanced image;

[0036] Figure 2 (b) is a schematic diagram of dark channel information of low-light enhanced image;

[0037] Figure 2 (c) is a schematic diagram of the reference image;

[0038] Figure 2 (d) is a schematic diagram of the dark channel information of the reference image. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] Example

[0041] like Figure 1 As shown, this embodiment provides a dark channel low-light image enhancement quality evaluation method, including but not limited to the following steps:

[0042] Input image: low light enhanced image x I and the reference image y I Enter as input image.

[0043] Dark channel information extraction: respectively for the input low-light enhanced image X I and the reference image Y I Extract the dark channel information and obtain the dark channel information X of the low-light enhanced image dark And the dark channel information Y of the reference image dark ,like Figure 2 shown.

[0044] Feature extraction: Transform low-light enhanced image X I Its channel information X dark , reference image Y I and its dark channel information Y dark Input into the CNN neural network to obtain the corresponding 5 overall features and dark channel features.

[0045] Calculate feature similarity: Calculate the similarity of the overall features and dark channel features of the low-light enhanced image and the reference image respectively.

[0046] Obtain the final quality score: Train and learn the scoring weights of the overall features and the dark channel feature similarity to obtain the final quality score.

[0047] The model used for dark channel information extraction is the dark channel extraction model, and the expression is as follows:

[0048]

[0049] Among them, J dark () represents dark channel information, J C () represents the haze-free image of channel C, C represents one of the three channels R, G, and B, and x represents the input image.

[0050] The CNN neural network extracts features from the dark channels of the low-light enhanced image and the reference image, and extracts the dark channel features. The feature extraction process is as follows:

[0051] The deep features corresponding to the dark channels of the low-light enhanced image and the reference image are calculated through the VGGNetCNN neural network;

[0052] In the VGGNetCNN neural network, the outputs of the five hidden convolutional layers are taken as the depth features corresponding to the dark channels of the low-light enhanced image and the reference image respectively;

[0053] Through the VGGNetCNN neural network, 5 dark channel depth features of the low-light enhanced image and the reference image are obtained respectively.

[0054] The CNN neural network extracts overall features of the low-light enhanced image and the reference image to extract general overall image features. The feature extraction process is as follows:

[0055] The deep features corresponding to the low-light enhanced image and the reference image are calculated through the VGGNetCNN neural network;

[0056] In the VGGNetCNN neural network, the outputs of the five hidden convolutional layers are taken as the overall features of the low-light enhanced image and the reference image respectively;

[0057] The VGGNetCNN neural network is used to obtain 5 overall features of the low-light enhanced image and the reference image.

[0058] The calculation formula for the similarity between the dark channel depth features and the overall features of the low-light enhanced image and the reference image is as follows:

[0059]

[0060]

[0061] Where i represents the i-th feature, i = 1, 2, 3, 4, 5, and Respectively represent the similarity between the i-th dark channel depth feature and the i-th overall feature, and They represent the i-th dark channel depth feature and the i-th overall feature of the low-light enhanced image, respectively. and They represent the i-th dark channel depth feature and the i-th overall feature of the reference image respectively.

[0062] According to the similarity between the dark channel depth feature and the overall feature, the quality score is calculated to evaluate the effect of the low-light enhancement algorithm. The final quality score obtained through training and learning is:

[0063] S=w dark D dark +w I D I

[0064] Among them, w dark and w I are the scoring weights of the dark channel depth feature similarity and the overall feature similarity obtained through training, D dark and D I They are the overall similarity of the dark channel depth features and the overall similarity of the overall features, namely:

[0065]

[0066] In summary, the embodiments of the present invention use dark channel information to obtain the illumination information of the image, and use the dark channel information and illumination information to evaluate the quality of the low-light enhanced image, which can produce better low-light enhanced image quality evaluation results. Dark channel information can be used as an evaluation index for image blur and interference, and can be used for low-light image enhancement quality evaluation to better determine whether blur and interference still exist after enhancement, and can better evaluate the quality of low-light enhanced images, and better assist in improving low-light image enhancement algorithms.

[0067] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A low-light image enhancement quality evaluation method based on dark channel, characterized in that: The low-light image enhancement quality evaluation method comprises the following steps: Input a low-light enhanced image and a reference image, and obtain the dark channels of the low-light enhanced image and the reference image through a dark channel extraction model; Using a CNN neural network to extract features of the dark channels of the low-light enhanced image and the reference image, and obtaining dark channel depth features of the low-light enhanced image and the reference image respectively; Using a CNN neural network to extract features from the low-light enhanced image and the reference image, respectively obtaining overall features of the low-light enhanced image and the reference image; comparing the similarity of dark channel depth features and overall features of the low-light enhanced image and the reference image; According to the similarity of the dark channel depth features and overall features of the low-light enhanced image and the reference image, the final quality score is obtained through training and learning.

2. The dark channel-based low-light image enhancement quality evaluation method according to claim 1, characterized in that: The dark channel extraction model expression is as follows: Among them, J dark () represents dark channel information, J C () represents the haze-free image of channel C, C represents one of the three channels R, G, and B, and x represents the input image.

3. The dark channel-based low-light image enhancement quality evaluation method according to claim 1, characterized in that: The CNN neural network extracts features from the dark channels of the low-light enhanced image and the reference image, and the process is as follows: The deep features corresponding to the dark channels of the low-light enhanced image and the reference image are calculated through the VGGNetCNN neural network; In the VGGNetCNN neural network, the outputs of the five hidden convolutional layers are taken as the depth features corresponding to the dark channels of the low-light enhanced image and the reference image respectively; Through the VGGNetCNN neural network, 5 dark channel depth features of the low-light enhanced image and the reference image are obtained respectively.

4. The dark channel-based low-light image enhancement quality evaluation method according to claim 1, characterized in that: The CNN neural network extracts features from the low-light enhanced image and the reference image, and the process is as follows: The deep features corresponding to the low-light enhanced image and the reference image are calculated through the VGGNetCNN neural network; In the VGGNetCNN neural network, the outputs of the five hidden convolutional layers are taken as the overall features of the low-light enhanced image and the reference image respectively; Through the VGGNetCNN neural network, 5 overall features of the low-light enhanced image and the reference image are obtained respectively.

5. The dark channel-based low-light image enhancement quality evaluation method according to claim 1, characterized in that: The calculation formula for the similarity between the dark channel depth feature and the overall feature of the low-light enhanced image and the reference image is as follows: Where i represents the i-th feature, i = 1, 2, 3, 4, 5, and Respectively represent the similarity between the i-th dark channel depth feature and the i-th overall feature, and They represent the i-th dark channel depth feature and the i-th overall feature of the low-light enhanced image, respectively. and They represent the i-th dark channel depth feature and the i-th overall feature of the reference image respectively.

6. The dark channel-based low-light image enhancement quality evaluation method according to claim 5, characterized in that: According to the similarity between the dark channel depth feature and the overall feature, the final quality score is obtained through training and learning as follows: S=w dark D dark +w I D I Among them, w dark and w I are the scoring weights of the dark channel depth feature similarity and the overall feature similarity obtained through training, D dark and D I They are the overall similarity of the dark channel depth features and the overall similarity of the overall features, namely:

Images