A Video Image Saliency Detection Method Based on Field Quantity Analysis

Abstract

The invention discloses a video image saliency detection method based on field quantity analysis, which comprises the following steps of: S1, obtaining a static saliency map of the video image; S2, extracting an optical flow vector field of the scene according to continuous video frames; S3, preliminarily classifying the optical flow vector field by a clustering method, and finding out the maximum classification block; S4, generating difference energy according to the contrast between each classification block and the maximum classification block; S5, standardizing the difference energy to obtain a dynamic saliency value and generating a dynamic saliency map; and S6, adding linear weighting of the dynamic saliency map with that of the static saliency map to obtain a final saliency map, thereby detecting the saliency of the video image. The method disclosed by the invention comprehensively utilizes static features and dynamic features of the video scene to obtain a saliency mapping result, and particularly analyzes the dynamic features of an object by the optical flow field quantity analysis to recognize important objects with definite motion features in the scene.

Description

technical field [0001] The invention belongs to the technical field of video image processing, and in particular relates to a video image saliency detection method. Background technique [0002] Recognizing important objects from complex scenes is a fundamental function of the human visual nervous system. For example, traffic lights can attract the attention of the human eye while driving, an airplane flying in the blue sky can attract the attention of the human eye, and a lighthouse at sea level can attract the attention of the human eye at night. Relying on this function, we can focus on the key parts to achieve better analysis results. [0003] Saliency detection is to enable the computer system to imitate the attention mechanism of the human eye, and highlight the important parts of the video image through the corresponding calculation process, which is a "discovery" process. Using the results of saliency detection, various scarce resources can be allocated preferentia...

AI Technical Summary

Problems solved by technology

[0007] Wixson et al. proposed a directional constant flow detection method in 2000, but it assumes that the target moves along a straight line, which is difficult to adapt to most application scenarios

Mahadevan et al. proposed a center-surround spatiotemporal saliency detection method in 2010. Its results strongly depend on the size of the detection window, and it is prone to detection failures for larger foreground objects.

Gopalakrishnan proposed a motion saliency detection method for linear dynamic contours in 2012. This method can only perceive the general position of the target, cannot generate a complete contour, and has poor accuracy.

Images