Video saliency detection algorithm based on saliency target background modeling

Abstract

The invention discloses a video saliency detection algorithm based on saliency target background modeling. The algorithm comprises steps as follows: step 1, performing static image saliency detection on an initial frame (current frame) to acquire a saliency map of the initial frame, and acquiring a residual background according to the saliency map; step 2, performing similarity filling on pixels with pixel values being 0 in a residual background image to acquire an initial frame background model; step 3, according to the initial background model, subtracting the initial background model from the next frame through background subtraction to acquire a next frame of salient zone, performing information redundancy distinguishing on the salient zone, if the redundancy is lower, judging the salient zone to be correct, if the redundancy of the salient zone is higher, performing background modeling again on the current frame with the methods in steps 1 and 2; finally, acquiring a final salient video through sequential circulation. According to the algorithm, background modeling is not required to be performed on each frame in the video, so that the calculation speed can be higher.

Description

technical field [0001] The invention relates to the technical field of computer vision and multimedia processing, in particular to a video saliency detection algorithm based on salient object background modeling. Background technique [0002] In recent years, with the development of research on content-based video relocation, it is required that the relocated video can be displayed in various screen sizes and retain as much important information as possible in the video. In these content-based video relocalization techniques, how to quickly and accurately detect visually salient regions is still a challenging problem to be solved. [0003] Traditional video saliency methods based on background modeling, such as the mixed Gaussian model background modeling method and the video target detection algorithm based on quaternion Fourier transform, need to combine the initial multi-frame images in the video for background modeling, so they cannot The frame image is detected, and be...

AI Technical Summary

Problems solved by technology

[0003] Traditional video saliency methods based on background modeling, such as the mixed Gaussian model background modeling method and the video target detection algorithm based on quaternion Fourier transform, need to combine the initial multi-frame images in the video for background modeling, so they cannot The frame image is detected, and because it adds inter-frame related information, there will be a disadvantage of high modeling complexity

The salient image quality determination rules for each frame are not introduced, so the salient video quality cannot be effectively guaranteed

Images