本論文所提出智慧型彩色影像紋理辨識技術,首先將彩色影像轉換為R、G、B三張影像,再將所獲得的三張影像利用奇異值分解加強影像之紋理並結合離散小波轉換特徵抽取方法抽取紋理色彩特徵,接著使用支持向量機及粒子族群最佳化演算法挑選最佳特徵組合與支持向量機參數。根據實驗結果顯示,本研究所提出之方法其效能優於其他現存的影像辨識方法。
This paper presents a novel intelligent color image texture recognition technique which is called the ICITR technique. First, it applies singular value decomposition (SVD) to each channel image of a RGB color image to enhance image textures. Subsequently, it extracts color texture features in discrete wavelet transform domain of the original channel image and its corresponding enhanced texture image with SVD. The ICITR technique employs a support vector machine (SVM) to be a multiclassifier. Meanwhile, particle swarm optimization is utilized to select the best feature combination and the parameters used in the SVM. The experimental results show that the ICITR technique can achieve satisfying results and also outperforms other existing methods under consideration here.