CN101867704A - Method for removing block noise from video image - Google Patents

Abstract

The invention relates to a method for removing block noise from a video image, which comprises the following steps of: performing entropy decoding, reordering, inverse qualification, inverse transformation and filtration with a filter on the video image to obtain a reconstructed image, and performing nonlinear diffusion image filtering processing on the reconstructed image. The step of performing nonlinear diffusion image filtering processing comprises the following steps of: firstly, calculating the gradient absolute value of each pixel in the image and the isolux curvature absolute value of the image and determining the diffusion coefficient value of each pixel according to the two parameters; secondly, performing nonlinear diffusion algorithm to remove the block noise, namely determining the degree of diffusion according to the diffusion coefficient and updating the gray-scale value of the pixel at the same time; and finally calculating the signal to noise ratio of the image diffusion result, if the signal to noise ratio is higher than a given value, finishing the image processing, otherwise, returning to the last step, namely updating the gradient absolute value and the curvature absolute value until the signal to noise ratio is higher than the given value. The method of the invention is suitable for filtering the block noise from the reconstructed image, can protect edges and has very good effect.

Description

A kind of method of removing the block noise of video image

Technical field

The present invention relates to a kind of method of removing the block noise of video image, belong to image processing and technical field of computer vision.

Background technology

In the H264/MPEG-4AVC video encoding standard, the encoding and decoding inverse transformation quantizes the back image block distortion can occur.The reason of its generation has two, a wherein important reason is in the block-based frame and DCT (discrete cosine) conversion of Inter prediction residue, the quantizing process of its conversion coefficient is coarse relatively, thereby the conversion coefficient that the inverse quantization process is recovered has error, can cause in the borderline vision of image block discontinuous; Secondly, Another reason comes from motion compensated prediction, motion compensation block may be never to be to duplicate in the interpolation sampling point data on the diverse location of same frame and come, because the coupling of motion compensation block can not be an absolutely accurate, so it is discontinuous to produce data on the border of copy block.

In H264, adopted loop filter to eliminate block distortion, but on the DCT border, just in time be the border of image, do not thought by mistake be block distortion, then may cause new error if do not judge.So when the filtering block distortion, need to judge that earlier this border is the real border or the formed border of block distortion of image.Real border is not carried out Filtering Processing, then will be to false border according to the character of image block and the filtering that varying strength is adopted in coding method on every side.To analyze the sample value that each needs filtered both sides, border in order to distinguish true and false border, threshold value is set then judges whether this sampling point is filtered, the result shows that the utilization loop filter can obviously alleviate block distortion in H264, but inevitably still there is block distortion, not enough fairing at the edge of image place.In order further to remove block distortion, can carry out post-filtering to the data in the display buffer, improve the subjective quality of image.

In the denoising smoothing process of image, linear diffusion is equivalent to gauss low frequency filter, it does not weaken all radio-frequency components of image with making any distinction between, thereby in level and smooth, blured the edge, therefore need find a kind of method of diffusion detected image edge automatically, thereby the coefficient of conductivity of diffusion process diminishes automatically near the important edges of image, even is close to zero.Nineteen ninety Perona and Malik have proposed famous P-M nonlinear diffusion equations, protection edge when it can be level and smooth, but the other side's block edge is handled and is had certain restriction.Therefore, H264 go whether can introduce a kind of nonlinear diffusing filter as post-filtering after the square loop filtering, further alleviate blocking artifact under the situation that the edge is protected, be still waiting research and development.

Summary of the invention

The purpose of this invention is to provide a kind of method of removing the block noise of video image, in image processing, effectively remove block distortion, effectively improve picture quality.

The technical solution adopted for the present invention to solve the technical problems is: a kind of method of removing the block noise of video image, at first video image is carried out that entropy is decoded, reordered, inverse quantization, inverse transformation and filter process obtain reconstructed image, reconstructed image is carried out the Nonlinear Diffusion image filtering again and handle, described Nonlinear Diffusion image filtering is handled and is comprised the steps:

The gradient absolute value of each pixel must arrive the edge according to the gradient absolute calculation and stop the parametric function value in a, the calculating reconstructed image;

The curvature of each pixel calculates curvature driving parameters value according to curvature in b, the calculating image reconstruction;

C, to stop parameter value with the curvature driving parameters value that obtains and edge be diffusion coefficient, carries out the Nonlinear Diffusion computing, obtains new image value Wherein, i, j are the pixel coordinate of image, and n is an iterations;

D, press

With The difference of signal noise ratio (snr) of image is judged, if less than set point, then begins repetitive operation from the first step, if be equal to or greater than set point, then finishes the processing to video image.

Further: the set point in described steps d≤10 ^-3

The present invention is owing to introduced the curvature driving parameters and the edge stops parameter as diffusion coefficient; on the basis at protection edge, effectively weakened block distortion; in the filtering of reconstructed image, use, compared with prior art, can improve the subjective assessment quality of image effectively.

Description of drawings:

Fig. 1 is a kind of realization block diagram of removing the block noise of video image that the embodiment of the invention provides;

Fig. 2 is that a kind of that the embodiment of the invention provides removes in the H264 decoding in the block noise processed of video image, carries out the step block diagram that the Nonlinear Diffusion image filtering is handled.

Embodiment:

Below in conjunction with drawings and Examples the present invention is further described:

Referring to shown in Figure 1, it is a kind of method of removing the block noise of video image that present embodiment provides, at first video image is carried out that entropy is decoded, reordered, inverse quantization, inverse transformation and filter process obtain reconstructed image, again reconstructed image is carried out the Nonlinear Diffusion image filtering and handles.

Referring to shown in Figure 2, it is that a kind of that present embodiment provides removes in the H264 decoding in the block noise processed of video image, carries out the step block diagram that the Nonlinear Diffusion image filtering is handled, and comprises the steps:

The first step is calculated the gradient absolute value of each pixel, stops the value of function according to gradient absolute calculation edge:

The view data of supposing the display buffer district is u _{I, j}, then the gradient absolute value is:

$\left|{\▿u}_{i,j}\right|=|u_i-u_j|$

Then will

Value substitution edge stops function

Wherein the value of K is generally got between 1.3～1.5, obtains the edge and stops parameter.

Second goes on foot, and calculates the curvature parameters κ of each pixel, and its computing formula is:

$\mathrm{\κ}=\frac{u_{\mathrm{xx}}{u}_y^2-2u_x{u}_y{u}_{\mathrm{xy}}+u_{\mathrm{yy}}{u}_x^2}{{(u_x^2+u_y^2)}^{3/2}}$

Wherein, u _xBe the single order partial derivative of image on the x direction, u _XxBe the second-order partial differential coefficient of image on the x direction, u _yBe the single order partial derivative of image on the y direction, u _YyBe the second-order partial differential coefficient of image on the y direction, u _XyBe that image is earlier asked the single order partial derivative on the x direction, and then on the y direction, ask second-order partial differential coefficient, with κ value substitution curvature driving function f (| κ |)=| κ | ^p, among the p=2, calculate the curvature driving parameters, wherein single order and second dervative all adopt the centered difference approximate calculation.

The 3rd step stopped parameter with curvature driving parameters and edge and combines as diffusion coefficient, carried out the calculating of Nonlinear Diffusion then, obtained new image value:

$u_{i,j}^{n+1}=u_{i,j}^n$

$+\frac{\mathrm{\Δt}}{\left|{\mathrm{\η}}_s\right|}\underset{p\∈{\mathrm{\η}}_s}{\mathrm{\Σ}}f\left(\right|{\mathrm{\κ}}_s^n\left|\right)g\left(\right|{\▿u}_{s,p}^n\left|\right){\▿u}_{s,p}^n$

Subscript s in the formula, p are the image pixel coordinate, and η _sBe to be the image neighbours territory set at center with s, | η _s|=4.Δ t is the time interval of iteration, and general set point is 1.

In the 4th step, calculate With

Signal to noise ratio poor, and compare with set point, if less than set point, then begin repetitive operation from the first step, if be equal to or greater than set point, then finish processing to video image.

Among the present invention, introduced curvature driving parameters and edge and stopped parameter as diffusion coefficient, effectively weakened block distortion on the basis at protection edge, this uses in the filtering of reconstructed image, can well improve the subjective assessment quality of image.

Claims (2)

1. method of removing the block noise of video image, it is characterized in that: with video image carry out that entropy is decoded, reordered, inverse quantization, inverse transformation and filter process obtain reconstructed image, reconstructed image is carried out the Nonlinear Diffusion image filtering again and handle, described Nonlinear Diffusion image filtering is handled and is comprised the steps:

The gradient absolute value of each pixel must arrive the edge according to the gradient absolute calculation and stop the parametric function value in a, the calculating reconstructed image;

The curvature of each pixel calculates curvature driving parameters value according to curvature in b, the calculating image reconstruction;

C, to stop parameter value with the curvature driving parameters value that obtains and edge be diffusion coefficient, carries out the Nonlinear Diffusion computing, obtains new image value

Wherein, i, j are the pixel coordinate of image, and n is an iterations;

D, press

With

The difference of signal noise ratio (snr) of image is judged, if less than set point, then begins repetitive operation from the first step, if be equal to or greater than set point, then finishes the processing to video image.

2. a kind of method of removing the block noise of video image according to claim 1 is characterized in that: the set point in described steps d≤10 ^-3

Images