Unsupervised texture segmentation via wavelet transform

摘要: Abstract In this paper, a mechanism for unsupervised texture segmentation is presented. The approach based on the multiscale representation of discrete (dyadic) wavelet transform which can be implemented by fast iterative algorithm. For it generally difficult to determine number classes identified. proposed offers an circumvent problem. Our method utilizes set high-frequency channel energies characterize features, followed multi-thresholding technique coarse segmentation. coarsely segmented results at same scale are incorporated intea-scale fusion procedure. A fine then used reclassify ambiguously labeled pixels generated from step. Finally, determined inter-scale in multiple scales integrated. performance demonstrated several experiments synthetic images, natural textures Brodatz's album and real-world textured images. Since choice wavelets very extensive open, we further explore various types time cost also measured.