Saliency Detection Algorithm Based On Space Comparison And Analysis Of Visual Feature

With the high speed development of information technology, images have becomethe primary data source of the society. Traditional manual image processing is confrontedwith many difficults because of the amount of images.This makes intelligent informationprocessing become the focus of attention.We notice that the interested content is only asmall part of the original image in the image processing such as image compression andimage retrieva. Therefore, it is important to import the salient region detection in imageprocessing in order to increase the effectiveness of information processing. In view of this,some visual attention computational model are proposed.A bayesian salient object extraction model based on space distribution and localcomplexity of the transition window is proposed. First of all, the bright saliency value mapis obtained by computing the contrast of local area and its multiple scales neighborhoodand then the color saliency value map is computed by using space and local feature ofcolor information. Meanwhile, the orientation saliency value map is obtained bymulti-scale analysis responses of Gabor filters. The above saliency values are inputted intothe single-scale Bayesian framework model based on transition-sliding window. Then theprobability of that a pixel’s salient is computed by comparing the saliency values insidethe window and outside the transition window. Finally, the saliency map of the inputimage is obtained by taking the maximum value, so the salient object is located andextracted according to the saliency map. The better test results show that the algorithm isfeasible and valuable.A random walk saliency visual attention algorithm based on global contrast isproposed. Firstly, the feature vector obtained by computing the color and orientationglobal contrast values is used to determine the weight of edge, and then the isolated regionis obtained by using the random walk on a complete graph to extract the global propertiesof the image. Meanwhile, the uniform region is enhanced by using the random walk on ak-regular graph to extract the local properties of the image. Finally, the saliency map isobtained by combining the global properties and local properties of the image, and the salient object is located and extracted according to the saliency map. Experimental resultsshow that the proposed algorithm is more reasonable and effective.A region contrast saliency estimation algorithm based on color and orientationcontrast of the segmented regions is proposed. First of all, the input image is segmentedinto regions using the algorithm of graph-based image segmentation, then define the colorsaliency for each region as the weighted sum of the region’s contrasts to all other regionsin the image. Meanwhile, we segment the image into regions using the algorithm oftexture segmentation. Then compute the orientation saliency for each region as theweighted sum of the region’s contrasts to all other regions in the image. Finally, thesaliency map of the input image is obtained by combining color saliency map andorientation saliency map, so the salient object is located and extracted. The results showthat the proposed algorithm is more reasonable and effective.