Gradient similarity video quality evaluation method based on low-frequency significance

Abstract

The invention relates to a gradient similarity video quality evaluation method based on low-frequency significance, which comprises the following steps of performing denoising processing on a video, decomposing the video into a prediction part and a noise part; performing video quality evaluation on the noise part to obtain a video quality score of the noise part; performing pixel-level gradient video quality evaluation, block grade video quality evaluation and visual significant attention video quality evaluation on the prediction part respectively, so as to obtain a video quality score of the prediction part; and obtaining a final quality evaluation score through synthesis of the video quality score of the noise part and the prediction part. By optimizing noise removal processing, the influence of random noise on video evaluation is effectively reduced. The invention provides an image content distortion and brightness distortion assessment simulation scheme, a visual salience attention similarity evaluation scheme and an overall video quality evaluation scheme. Compared with a traditional method, the gradient similarity video quality evaluation method in the invention has higherconsistency with a human visual system evaluation system.

Description

Technical field [0001] The invention relates to a gradient similarity video quality evaluation method based on low-frequency saliency, and belongs to the technical field of computer vision image processing. Background technique [0002] With the rapid development of multimedia technology and the gradual deepening of machine learning, humans’ way of observing the world is no longer limited to their own senses, and has developed to be able to use computers and cameras instead of human eyes to identify, track, and measure targets. , And further do image processing, so that computer processing becomes more suitable for human eyes to observe or send to the instrument to detect images. In the context of this technological development, image recognition is now a popular technology, and the evaluation technology for image quality is the core technology in image recognition. [0003] Scholars have discovered that the information processing of the human visual system is hierarchical, and th...

AI Technical Summary

Problems solved by technology

[0004] Problem 1: There is a lack of simulation research on image brightness distortion. Perceptual distortion can be divided into content-dependent distortion and content-independent distortion. The change of pixel brightness in the image area, that is, the gradient of the image reflects the content of the image, but the lack of image brightness distortion will affect the overall video. Some loss in accuracy of quality similarity assessment

[0005] Problem 2: There is a lack of image filtering methods with high matching adaptability. Traditional filtering methods have limitations in processing mixed noise, and it is difficult to obtain good filtering effects.

[0006] Problem 3: There is a lack of detailed research on the saliency detection of reference images and test images. When judging salient pixel values, the imprecise value of salient values ​​will interfere with the overall video quality assessment

Images