Multiresolution adaptive and progressive gradient-based color-image segmentation

bstract. We propose a novel unsupervised multiresolution adapive and progressive gradient-based color-image segmentation algoithm (MAPGSEG) that takes advantage of gradient information in n adaptive and progressive framework. The proposed methodolgy is initiated with a dyadic wavelet decomposition scheme of an rbitrary input image accompanied by a vector gradient calculation f its color-converted counterpart in the 1976 Commission Internaionale de l’Eclairage (CIE) L*a*b* color space. The resultant gradint map is used to automatically and adaptively generate thresholds o segregate regions of varying gradient densities at different resoution levels of the input image pyramid. At each level, the classifiation obtained by a progressively thresholded growth procedure is ntegrated with an entropy-based texture model by using a unique egion-merging procedure to obtain an interim segmentation. A condence map and nonlinear spatial filtering techniques are combined, nd regions of high confidence are passed from one resolution level o another until the final segmentation at the highest (original) resoution is achieved. A performance evaluation of our results on sevral hundred images with a recently proposed metric called the noralized probabilistic Rand index demonstrates that the proposed ork computationally outperforms published segmentation techiques with superior quality. © 2010 SPIE and IS&T. DOI: 10.1117/1.3277150