Accurate target segmentation method based on color significance and Gaussian model

An accurate target segmentation method based on color significance and a Gaussian model is disclosed. The method is characterized by firstly, clustering image pixels in a Lab color space through a GMMalgorithm; then, using a SSIM image similarity algorithm to merge sub-Gaussian models, and selecting a target sub-Gaussian model as a foreground through prior color information; and then, using a CRFalgorithm to optimize a significance region and obtaining an accurate segmentation boundary. In the invention, aiming at a characteristic that an object may not accord with center and boundary priorduring significance detection, a significance object detection method based on color prior is provided. The pixels are clustered directly through the Gaussian mixture model and the center and boundaryprior is not used; the Gaussian mixture model is used to ensure that an accurate and stable boundary is acquired; the significance area located at the boundary can be detected; and compared with a traditional significance detec