集成多特征与稀疏编码的图像分类方法

摘要

采用单一特征时存在提取信息量不足、对图像内容描述较片面等问题，单一编码方法在组织特征向量时也会对图像造成过多的信息丢失。针对这些问题，文中提出一种集成多特征与稀疏编码方法。首先，对图像进行空间金字塔划分，结合尺度不变特征和梯度方向直方图特征之间的优势互补性，提取得到不同的特征集。然后，在不同的特征集上用不同的聚类方法得到不同的视觉词汇本，在每个词汇本上分别进行局部稀疏编码和稀疏编码，得到不同的图像描述集。最后，利用线性SVM进行图像分类，并对得到的多个结果采用投票决策方法决定最终分类情况。实验表明文中方法有良好的准确性和鲁棒性。%Using a single image feature to describe the image content is one-sided because of the insufficient information. Besides, the single coding method usually loses the spatial information. To solve these problems, an approach of integrating multi-features and sparse coding methods is proposed. Images are firstly divided into sub regions according to the spatial pyramid, and then the complementary advantages of scale invariant feature transform and the histogram of oriented gradients features are combined to produce various feature sets. Then, different clustering methods are used on different feature sets to acquire different codebooks. Next, two sparse coding methods, locality constrained linear coding and sparse coding based on each codebook are further employed respectively to get various image description sets. Finally, linear support vector machines are applied to image classification, and a voting method is used to determine the final classification. Experimental results show that the proposed method has good accuracy and robustness compared with some state-of-the-art methods.