Context-aware Saliency Detection Algorithm For Graphics/Images And Their Applications

Saliency detection plays an important role in context-based processing of imagesand computer graphics. This paper presents some novel saliency detection algorithmsfor2D images and3D geometric models. Some applications of these saliencymeasurements are also given.For the saliency detection of2D image, we propose a new context-aware detectionalgorithm, which defines pixel saliency as the dissimilarity between the patch centeredat the current pixel and the patches centered at “other pixels”. We also gives a novelstrategy to speed up the saliency computation scheme, that is, by confining thecomputational range of the above “other pixels” and adding the histogram informationto eliminate the loss of accuracy. Owing on our saliency detection algorithm, we alsopropose a saliency-based bilateral filtering algorithm and a saliency-based edgedetection algorithm. Our saliency-based bilateral filtering can preserve the importantregions of the underlying images by weakening the filter strength at these areas. Oursaliency-based edge detection algorithm can erase those unimportant edges from thedetected edge map. Combined the extracted edges with the filtered image, we can obtaina stylized rendering technique.For the saliency detection of3D models, a perceptual-saliency measure isfirstly introduced, which is defined by a center-surround bilateral filter operator onGaussian-weighted average of local projection height between mesh vertices andtheir neighbors. Owing to the perceptual-saliency measure and saliency-domainshape modeling techniques, a novel shape manipulation scheme is then presented.The geometric features of the underlying shapes can be filtered by low-pass,high-pass and enhancement due to the user specified shape manipulation functionsdefined on the saliency-domain, that is, the low-pass shape manipulation, thehigh-pass shape manipulation and the enhancement shape manipulation.