根據空間-頻率域分析之紋理切割方法

Abstract

在我們的論文中,我們採用了一套雙視窗的空間-頻率域分析方法,藉由此方法量測空間中不同區域的頻率資訊,而量測到的頻率資訊則當成紋理特徵用以處理紋理切割的問題.雙視窗空間-頻率分析法在空間和頻率域上各有一個觀察資訊用的視窗,空間視窗選取空間中分析的區域範圍,並決定了紋理資訊在頻率域上的解析度,頻率域視窗則挑選合適的頻段,藉由結合頻段內的資訊取得紋理特徵.此外,現有的空間-頻率域分析法一般會量得震盪劇烈的紋理特徵,以致產生不正確的切割結果.在本論文中,我們也針對紋理特徵產生震盪的原因加以分析,並經由一些處理以減少紋理特徵震盪的現象.

In this thesis, we present a new feature based texture segmentation scheme using a dual-window spatial-frequency domain approach. The dual-window spatial-frequency domain approach has two observing windows, one for spatial domain and the other one for frequency domain. The spatial domain window is used to choose the observed region and decide the resolution of texture information in the frequency domain. The frequency domain window is used to mix up observed frequency components to form texture features. In addition, the oscillating magnitude of texture features is also discussed in detail. Using texture features with large oscillation will make the segmentation more difficult. In the thesis, the causes of the unwanted oscillation is analysed systematically and some methods are proposed to suppress the oscillation.