CN108416770B - Image quality evaluation method based on visual saliency - Google Patents

Abstract

The invention provides an image quality evaluation method based on visual saliency, which relates to the technical field of image processing. The method has the advantages that the visual saliency concept is introduced into the image quality evaluation, and the visual saliency interfered in the image is described by analyzing the distribution uniformity of the interference in the image, so that the subjective influence degree of the interference in the image on human vision is reflected, and the image quality is evaluated more accurately.

Description

Image quality evaluation method based on visual saliency

Technical Field

The invention relates to the technical field of image processing, in particular to an image quality evaluation method.

Background

The image quality evaluation is one of basic technologies in the field of image processing, and plays an important role in aspects of algorithm analysis comparison, system performance evaluation and the like, and the full-reference image quality evaluation is a method for evaluating the quality of a distorted image by comparing the difference between the distorted image and an original image, and is widely applied to the fields of evaluating image coding compression, guiding image processing algorithms, monitoring image quality of user terminals and the like.

On the basis of the traditional image quality evaluation method, the subjective perception difference of people is described by edge features by using the perception characteristic of a visual system of human eyes which is extremely sensitive to image edges, and the method only has higher performance in the evaluation of a few compression algorithms and is still inaccurate in subjective perception description of human eyes on the whole and lacks subjective consistency.

Disclosure of Invention

In order to overcome the defects of the prior art, the invention provides the method for comprehensively evaluating the image quality by describing the significance of interference through the density degree of local structure change and combining the local structure change intensity, the change total amount, the change pixel and the contour line relationship so as to improve the evaluation accuracy.

The technical scheme adopted by the invention for solving the technical problem comprises the following specific steps:

step 1, respectively converting an image to be detected and a reference image into a gray image I_distAnd I_refFor two gray scale images I_distAnd I_refRespectively calculating phase consistency PC under frequency domain_distAnd PC_ref；

Step 2, the gray gradient map calculation formula is as follows:

wherein G represents a gradient map, d_xAnd d_yGradient templates in both transverse and longitudinal directions:

wherein, a₁And a₂Satisfies the following conditions:

i is a gray image, and the gray image I can be obtained according to the formula (1)_distAnd I_refGray scale gradient map G_distAnd G_ref；

Step 3, calculating two gray gradient images G_distAnd G_refSimilarity of gradients between S_GGradient similarity map S_GValue S of each pixel point p in (1)_G(p) is:

G_dist(p) and G_ref(p) respectively represent FIG. G_distAnd G_refThe gray value of the middle pixel point p; wherein, T₂Is greater than 0, so as to obtain the weighted similarity S of two gray gradient images_GPCComprises the following steps:

wherein, max (PC)_dist,PC_ref) Is PC_distAnd PC_refOf the maximum value of the corresponding element, S_GDetermined by the calculation of formula (2);

step 4, a gradient similarity graph S is subjected_GThe distribution uniformity S is obtained_D：

First, the gradient similarity map S_GDown-sampling, dividing it into n square blocks B with even side length_iWhere i is 1,2,3 … n, each block B_iThe maximum value of the gray value of the middle pixel point is used as the gray value of the pixel point at the corresponding position after down sampling, so that a down sampling image S is obtained_G(max)Of size n_x×n_yAnd the gray value of each pixel point in the down-sampling graph is p_x,yAnd x and y indicate that the pixel is at S_G(max)The horizontal coordinate and the vertical coordinate in the step (2) are used for calculating the distribution uniformity S_D：

S_D＝|S₁+S₂|+|S₃+S₄|

Wherein S₁,S₂,S₃,S₄Intermediate variables are:

and 5: calculating mass fraction

Wherein 0 < sigma < 10, S_qualityNamely the quality evaluation score of the distorted image relative to the reference image.

The method has the advantages that the visual saliency concept is introduced into the image quality evaluation, and the visual saliency interfered in the image is described by analyzing the distribution uniformity of the interference in the image, so that the subjective influence degree of the interference in the image on human vision is reflected, and the image quality is evaluated more accurately.

Drawings

FIG. 1 is a flow chart of the present invention.

Detailed Description

The invention is further illustrated with reference to the following figures and examples.

The human visual perception system is more sensitive to high-frequency information, so that the significance degree of the change of the spatial structure of image content caused by various disturbance factors is undoubtedly the core factor influencing the subjective perceptibility of image quality. At present, the existing full-reference image quality evaluation method only starts from image difference analysis, and does not pay attention to the concept of significance.

The image quality evaluation method based on visual saliency of the invention has a flow chart as shown in fig. 1, and the specific implementation method comprises the following steps:

step 1: converting the original reference image and the image to be measured obtained by distortion processing into a gray image I_distAnd I_refFourier transform is carried out on the two images and phase consistency PC is calculated_distAnd PC_ref。

Step 2: an image gradient map is calculated. Defining transverse and longitudinal gradient templates d_xAnd d_yThe following were used:

convolution of an image with a gradient template to compute a gradient map

When I is taken out of I respectively_distAnd I_refRespectively obtain G_distAnd G_ref。

And step 3: calculating a weighted similarity S_GPC. First, the gradient similarity S is calculated_GFor each pixel point p in the gradient map, the similarity is obtained

Wherein T is taken out₂＝160。

Then, the weighted similarity S is calculated_GPC：

Where max (PC)_dist,PC_ref) Is PC_distAnd PC_refThe maximum value of the corresponding element in the matrix.

And 4, step 4: the distribution uniformity S is obtained_D。

Firstly, similarity graph S_GDown-sampling, continuously dividing the image into a plurality of blocks with the size of 8 multiplied by 8 pixels, taking the maximum value of the gray value of the pixel point in each block as the gray value of the pixel point at the corresponding position in the down-sampling image, and obtaining the down-sampling image S_G(max)Of size n_x×n_yThe gray value of each pixel point in the graph is p_x,_yAnd x and y indicate that the pixel is at S_G(max)The horizontal coordinate and the vertical coordinate in the step (2) are used for calculating the distribution uniformity S_D：

S_D＝|S₁+S₂|+|S₃+S₄|

Wherein S is₁,S₂,S₃,S₄Intermediate variables are:

and 5: calculating a quality assessment score

Wherein σ is 4, S_qualityNamely the quality evaluation score of the distorted image relative to the reference image.

Claims (1)

1. An image quality evaluation method based on visual saliency is characterized by comprising the following steps:

step 1, respectively converting an image to be detected and a reference image into a gray image I_distAnd I_refFor two gray scale images I_distAnd I_refRespectively calculating phase consistency PC under frequency domain_distAnd PC_ref；

Step 2, the gray gradient map calculation formula is as follows:

wherein G represents a gradient map, d_xAnd d_yGradient templates in both transverse and longitudinal directions:

wherein, a₁And a₂Satisfies the following conditions:

i is a gray image, and the gray image I can be obtained according to the formula (1)_distAnd I_refGray scale gradient map G_distAnd G_ref；

Step 3, calculating two gray gradient images G_distAnd G_refSimilarity of gradients between S_GGradient similarity map S_GValue S of each pixel point p in (1)_G(p) is:

G_dist(p) and G_ref(p) respectively represent FIG. G_distAnd G_refThe gray value of the middle pixel point p; wherein, T₂Is greater than 0, so as to obtain the weighted similarity S of two gray gradient images_GPCComprises the following steps:

wherein, max (PC)_dist,PC_ref) Is PC_distAnd PC_refOf the maximum value of the corresponding element, S_GDetermined by the calculation of formula (2);

step 4, a gradient similarity graph S is subjected_GThe distribution uniformity S is obtained_D：

First, the gradient similarity map S_GDown-sampling, dividing it into n square blocks B with even side length_iWhere i is 1,2,3 … n, each block B_iThe maximum value of the gray value of the middle pixel point is used as the gray value of the pixel point at the corresponding position after down sampling, so that a down sampling image S is obtained_G(max)Of size n_x×n_yAnd the gray value of each pixel point in the down-sampling graph is p_x,yAnd x and y indicate that the pixel is at S_G(max)The horizontal coordinate and the vertical coordinate in the step (2) are used for calculating the distribution uniformity S_D：

S_D＝|S₁+S₂|+|S₃+S₄|

Wherein S₁,S₂,S₃,S₄Intermediate variables are:

and 5: calculating mass fraction

Wherein 0 < sigma < 10, S_qualityNamely the quality evaluation score of the distorted image relative to the reference image.

Images