Abstract: In the image classification process, the classification accuracy is affected by the randomicity of feature point selection. In order to solve this problem, a self-learning classification algorithm based on image object feature space is proposed. The image target area is found with the local multi-channel active contour model based on the colour and texture feature. Features are selected in the target area and it makes sparse coding on features, which is used to establish the feature space. In order to further improve the classification accuracy, a voting mechanism is established on the proposed algorithm. Experimental results indicate that the proposed algorithm can avoid the impact of the randomicity of feature point selection, and effectively improve the accuracy of image classification.

Key words: Speeded Up Robust Features（SURF）, image segmentation, image classification, self learning, active contour model

摘要： 为解决图像分类过程中特征点选择的随机性对分类精度造成的影响，提出一种基于图像目标特征空间自学习分类算法。利用基于颜色和纹理特征的多通道局部主动轮廊模型找到图像的目标区域，在目标区域选取特征并对特征稀疏编码建立图像的目标特征空间。为进一步提高图像分类精度建立投票机制下基于图像目标特征空间的自学习算法。实验结果表明，该方法能避免特征选择的随机性对实验结果的影响，有效地提高图像分类的精度。

关键词: 加速稳健特征（SURF）, 图像分割, 图像分类, 自学习, 主动轮廊模型