Saliency-based Image Compression And Video Saliency Detection

Standards of JPEG2000have more applications in recent years for itscharacteristics of high compression rate, strong ability of resistance toerror and progressive transmission. JPEG2000based Region of interest(ROI) coding is one of them. There are two most important problem in it.One is to get the interested region, other is how to design weightingfunction makes a given compression rate win the best visual effects.Therefore, studies on how to get the interested region and constructweight function has important practical significance for the improvementof still image compression standard JPEG2000.This paper applys methods of Saliency detection in computer visionto obtain the interested region, puts forward a quick and accurate methodof obtaining region of interest. Based on thorough research on region ofinterest coding of JPEG2000, this paper proposes a JPEG2000basedSalient region compression algorithm JSRC2000. This algorithm usingthe results of saliency detection to constructure a JPEG2000compressionweighting function, and put this weighting function to JPEG2000compression algorithm. It can largely improve region of interest codingof JPEG2000compression algorithm. Meanwhile, this paper proposes anew prediction based video significant extraction method PVSD(predictive video saliency detection), which is mainly dividing a videointo different frame groups and using frame prediction mode to calculate the saliency map of each frame in the group.In order to validate efficiency of our algorithms, the paper carrys on avarious comparison and analysis with domestic and foreign latest severalalgorithms. Theory analysis and simulation experiment results show thatthe presented algorithms have higher efficiency than all other methodsand have better coding performance.