融合纹理信息的SLIC算法在医学图像中的研究

随着超像素算法的发展，SLIC（Simple linear iterative clustering）由于时间复杂度低及良好的分割结果而被广泛关注.但是由于传统的SLIC算法并没有考虑到图像的纹理信息，故而对于纹理较复杂的图像分割效果略有不足.LBP（Local binary pattern）对于纹理的识别有着优秀的表现而且时间复杂度低，但是对于噪声的鲁棒性较差，并且会产生纹理偏移.因此，本文首先针对传统的LBP中存在的问题进行改进；然后将改进后的算法与SLIC结合，提出一种融合纹理信息的超像素算法——SLICT（Simple linear iterative clustering based on texture）.为验证分割效果，本文选取纹理较多的医学图像进行实验，采用心脏MRI数据库进行验证并与其他超像素算法进行对比.实验表明，SLICT在边缘召回率、欠分割错误率以及覆盖率上的综合表现优于其他算法.从分割结果上来看，SLICT不但能够更好地贴合图像边缘，而且对于连续区域的分割效果也较好，更适合纹理较复杂的图像.

SLICT: Computing Texture-Sensitive Superpixels in Medical Images

With the development of superpixel algorithm, SLIC (Simple linear iterative clustering) is widely noticed because of its low time-complexity and remarkable results. However, traditional SLIC does not take texture information into consideration, which leads to unsatisfactory results on texture-complex images. LBP (Local binary pattern) algorithm has good performance on texture analysis. But it is not robust enough to texture noise, and it may also lead to texture deviation. So, we propose simple linear iterative clustering based on texture (SLICT) by improving the traditional LBP algorithm and combining it with SLIC. To verify our method's effectiveness, we use medical images which contain complex texture from cardiac MRI dataset and compare our algorithm with other superpixel algorithms. The experiment results show that SLICT has better performance than the other algorithms in boundary recall, under-segmentation error and coverage rate. Besides, image segmentation results show that SLICT can adhere to boundary more tightly and have better results on consistent area, which proves SLICT is more suitable to texture-complex images.