運用視覺模型和碎形波所作的紋理切割

Abstract

在影像分析中, 影像切割是一個基礎而重要的步驟, 由於紋理影像的 特殊結購, 紋理影像, 至今仍是具挑戰性而未解決的工作, 古典的影像處 理方法無法作出成功的紋理切割, 但是，人類卻視覺可以. 於是我們根據 之前的研究, 使用一個新的早期視覺模式來模擬人類視覺, 將一張紋理影 像轉換成距離地圖的表示法, 以突顯出不同紋理的邊界. 因此, 紋理切 割將變得可行. 我們進一步利用碎形波來描寫邊界函數的尖角和平滑 部分, 修改能 量函數來反應紋理的特性; 能量函數的最優化可以運用碎 形波的多重解析法加數計算過程, 同時加上模擬冷凝法以求得整體最優 化. 由於影像的表示法及分割技術的進步, 我們可以成功的從不同的紋理 影像中切割出紋理的邊界. Image segmentation is a fundamental important step for image analysis. Tremendous efforts have been made to develop roboust and efficient segmenation techniques previously. However, segmentation for texture images remains as a chalenging and unresoved problem due to its textural feature. While classical approaches may fail to give successful segmentation for texture images , humanvision demonstrates its incredile ability in localizing the boundaries among various textures. Encouraged by the human visual performance, a new early vision model been proposed in one of our pervious works attempting to mimic the human visual perception . This model converts a texture image into a new representation called distance map. Since the boundaries of textures are highlighted in the distance map, segmentation of a texture image turns to be manageable. Based on the new vision model, the boundary function is descibed by the spatially adaptive wavelets that can represent the sharp corners as well as smooth boundaries. New global energy functions built on local texture properies are investigated. The optimization is carried out via the full power of multiresolutionanalysis in wavelets. Simulated annealing is applied at the same time to find the global optimization. As a result, both the improvements of image representation and segmentation technique have advanced the state of art in this area. Simulation and empirical studies on texture images have been performed. The results confirm the enlightenment.