A Hevc-Based Saliency Detection Method in Compressed Domain

Abstract

The present invention provides a HEVC-based compressed domain saliency detection method, which relates to the field of video technology processing. By extracting chroma, brightness and texture static feature maps, a model for filtering the background of the static feature maps is established to calculate the final static saliency map. and a dynamic saliency map to get the final saliency map. Compared with other saliency models, the accuracy of the present invention is improved to a certain extent, and the detection algorithm is relatively stable, and at the same time, the complexity of the algorithm is correspondingly reduced. Gaussian distribution is used to filter out the background of the static feature map, and the static feature information is fully extracted, so that the calculation accuracy is improved. ; The motion feature, texture feature and statistical properties of the image are included in the calculation of the dynamic saliency map, which reduces the complexity of the algorithm, because the saliency of the compressed domain reflects the saliency of the reconstructed video, so it is more conducive to perceptual video coding; An adaptive fusion algorithm is adopted to make the saliency value more accurate.

Description

technical field [0001] The invention relates to the field of video technology processing, in particular to a method for detecting the significance of a video sequence. Background technique [0002] Perceptual video coding takes the video quality perceived by human eyes as the standard of video coding, and is considered to be a new coding form that can solve the bottleneck of existing video coding. Visual saliency models can be applied to perceptual video coding. So far, many saliency detection models have been studied, and the models are divided into bottom-up and top-down mechanisms. The former refers to a fast processing method based on low-level visual features and data-driven, while the latter refers to a slow processing method based on task-driven and conscious domination. [0003] The original bottom-up saliency detection models are inspired by biological vision. The study found that there is a working mechanism in the retina and primary visual cortex: visual neuron...

AI Technical Summary

Problems solved by technology

Although a new research direction is proposed, the marked area is not obvious, and the accuracy of experimental detection is not high

[0004] The top-down selective attention mechanism of biological vision is an unresolved problem in cognitive psychology. In the field of machine vision, the top-down visual saliency detection model is also a research difficulty. The general idea of ​​the existing models is First learn the prior knowledge related to the target and task, and then use the learned prior knowledge to guide the saliency detection. Xu et al. established a data set for human eye tracking, and used the feature component CU in HEVC to divide the depth and bit allocation. And the motion vector, respectively solve the characteristic information of space and time, and then use the nonlinear part of the support vector machine to classify, and finally use the training results to classify each pixel. This algorithm is more complicated to calculate and takes a long time.

Most top-down saliency detection models need to learn from huge image databases, the amount of calculation is huge, and they are not universal because the driving tasks are not universal

Images