论文部分内容阅读
结合支持向量机SVM和马尔可夫随机场MRF在图像分类领域的优点,提出MRF-ν-SVC分类系统。与以往针对像素的分类不同,本系统在区域水平实现分类。首先通过双阈值准则判别区域间边缘的强弱,然后针对模糊边缘区域将基于MRF的空间语境模型改进为边缘语境模型,修正SVM原始问题中的偏差因子,从而优化最优分类超平面。实验表明,本文算法可有效提高分类精度,增强对斑点噪声的抑制能力,更好地分类特征相似的区域。
Combined with the advantages of SVM and Markov random MRF in the field of image classification, the MRF-ν-SVC classification system is proposed. Different from the past for pixel classification, the system at the regional level to achieve classification. Firstly, the threshold of inter-area is discriminated by the double-threshold criterion. Then, the spatial context model based on MRF is improved to the edge context model for fuzzy edge region, and the deviation factor in the original problem of SVM is modified to optimize the optimal classification hyperplane. Experiments show that this algorithm can effectively improve the classification accuracy, enhance the suppression of speckle noise and better classify the areas with similar characteristics.