An inter-frame image coding method based on spatio-temporal saliency fusion

Abstract

The invention discloses an inter-frame image encoding method based on spatio-temporal saliency fusion, relates to the technical field of image processing, and mainly includes the steps of: obtaining a temporal saliency map according to the time-domain motion vector of each pixel point; extracting a feature based on the mean value of each pixel point A single-layer image with superpixel features in the inter-frame image; obtain the transfer matrix according to the single-layer image and the mean value characteristics of the corresponding pixel points of the nodes; according to the transfer matrix, obtain the spatial saliency map based on the Markov chain theory; according to the time The weight relationship between the saliency map and the spatial saliency map is used to obtain the spatio-temporal saliency map; the saliency map is obtained according to the spatio-temporal saliency map; the quantization parameters are dynamically adjusted according to the mean value characteristics of the corresponding pixels in the saliency map, and the inter-frame images are processed according to the quantization parameters to encode. The invention combines motion features in time domain and space domain of the image to obtain a saliency map in time and space domain for encoding, so that the encoded data contains more image information, and the fidelity after decoding is improved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an inter-frame image coding method based on spatio-temporal saliency fusion. Background technique [0002] At present, with the application of H.265 / HEVC and its extended coding standards more and more widely, the known perceptual computing models are mainly divided into four categories: the region of interest computing model (Region of Interest, ROI), visual attention model ( Visual Attention), visual sensitivity model (Visual Sensitivity), cross-perceptual attention model (Cross-Modal Attention). The perceptual coding methods can be divided into three categories: preprocessing methods, non-scalable coding methods, and scalable coding methods. Among them, the preprocessing method usually performs visual optimization processing on the inter-frame images in the original video before encoding, and it does not need to make changes to the encoder. The non-scalable coding m...

AI Technical Summary

Problems solved by technology

Although the research on the saliency of static images has made great progress in the world, the research on the saliency detection of dynamic inter-frame images is still in its infancy and has not yet formed a system.

However, the existing perceptual coding framework is not perfect enough. The information interaction between the image sensor and the encoder between frames is limited to salient object detection, which is not conducive to the information sharing between the two (such as foreground and background division information, motion type information, etc. )

Images