Font Size: a A A

Research Of Image Compression Method Based On Region

Posted on:2010-12-21Degree:DoctorType:Dissertation
Country:ChinaCandidate:B B HeFull Text:PDF
GTID:1118360275455422Subject:Circuits and Systems
Abstract/Summary:PDF Full Text Request
Image compression is one of the most important branch of image processing, and has widespread use in many fields.With the development of the technology, image compression should not only have high compression rate,but also have better visual effects.Based on the image segmentation,the paper analyze the extraction of the region of interest,foreground extraction and image compression method based on the saliency map of the region.the detailed contents are as follows.Propose the image compression method based on region saliency map, compress the low frequency sub-image with low bit-stream rate and ROI in high frequency sub-image with high bit-stream rate,then form the compression bit-stream.The method can keep the edge and ROI information and impove the compression ratio.Propose the rapid RSST image segementaion method,class label distance is used to define the similarity and the count of region combination is reduced corresponding.The method can reduce the segementaion time and keep the region's uniformity.Based on graph-cut method,a foreground extraction method that can process the nosie area is proposed,form the new energy function with the intrdouction of punish item and guass model,the precise foregound object can be picked out completely.A test system of region based image compression is designed and realized. Encouraging experimental results from BSDS image database illustrate the validity of the proposed methods in this dissertation.Some basic research related with image compression based on region is accomplished,but many problems are needed to analyze and reserach deeply.For example,as the development of the visual model,how to choose the ROI better.
Keywords/Search Tags:image segmentation, region of interest, image segmentation, graph cut, recursive shortest spanning tree, foreground extraction, predictive content model
PDF Full Text Request
Related items