An accurate object segmentation method based on color saliency and Gaussian model

Abstract

An accurate target segmentation method based on color saliency and Gaussian model. First, the image pixels are clustered in the Lab color space by the GMM algorithm, and then the sub-Gaussian models are merged by the SSIM image similarity algorithm, and selected by the prior color information. The target sub-Gaussian model is used as the foreground, and then the CRF algorithm is used to optimize the saliency region to obtain accurate segmentation boundaries. Aiming at the characteristics that objects in the saliency detection may not conform to the center and boundary priors, the present invention proposes a saliency object detection method based on color priors. The present invention directly uses the Gaussian mixture model to cluster pixels, without using the center and boundary priors. The Gaussian mixture model is applied to ensure accurate and stable boundaries, which can detect saliency regions located on the boundaries, and have higher accuracy than traditional saliency detection algorithms.

Description

technical field [0001] The invention belongs to the technical field of computer vision and is used for segmenting salient objects in images, and is an accurate target segmentation method based on color saliency and a Gaussian model. Background technique [0002] The human visual attention mechanism has the function of selectively processing visual images. Introducing this mechanism into image processing has many applications, including target detection, salient region extraction, etc. This selective attention mechanism can make the visual information Processing is faster. Research on visual psychology shows that the attention mechanism includes two processes, namely, a fast bottom-up data-driven process and a slow top-down goal-driven process, which is similar to the human visual attention mechanism, and saliency detection The algorithm is also divided into bottom-up and top-down. The bottom-up predicts the visual gaze point model according to the local features of the imag...

AI Technical Summary

Problems solved by technology

In addition, the GC (global cues) algorithm proposed by Cheng M M [6] The Gaussian Mixture Model GMM (Gaussian Mixture Model) is used to detect salient objects, but GC uses GMM to divide the RGB color model, which has a poor effect on pixel clustering, and the obtained classification boundaries are blurred, and the sub-Gaussian model is merged bad in most cases

Classic GrabCut algorithm [7] The Gaussian model is also used to model the foreground and background images, and achieved good results, but the GrabCut algorithm requires the user to set the initial foreground and background areas

All of the above are general salient object detection algorithms. At present, there is no salient object detection algorithm for color prior. Based on color prior, the detection accuracy of salient objects can be improved.

Images