基于颜色匹配和改进LBP的胶囊内镜视频缩减

针对胶囊内镜视频中存在大量的冗余图像,提出了一种基于颜色匹配和改进LBP的视频缩减算法。该算法首先将胶囊内镜图像转换到HSI空间并从中提取色彩信息,通过余弦角对相邻两帧的色彩信息进行相似性度量来实现颜色匹配。接着构建图像金字塔,用改进的LBP算子提取出多尺度的纹理特征,然后根据相邻两帧纹理统计的加权曼哈顿距离和纹理不同区域面积占全局的比例来进行纹理的相似性度量。最后采用级联分类器完成对图像的分类,将差异图像组建成精简视频从而实现视频缩减。实验结果表明,算法得到的召回率,准确率和综合性能指标分别达到了0.91,0.87和0.89,处理速度达0.055 s/帧。该算法与已有的算法相比,分类效果更好并且处理速度更快。

Capsule endoscopy video reduction based on color matching and improved LBP

A capsule endoscopy video reduction algorithm based on color matching and improved LBP is proposed to solve the problem of a large number of redundant images in the video.First,the image is converted into HSI space, and the similarity of the feature is measured by cosine angle for color matching.Then,the image pyramid is constructed,from which multi-scale texture features are obtained by using the improved local binary pattern operator.The similarity measurement of the texture is conducted by computing weighted Manhattan distance of the texture feature between two successive frames and calculating the proportion of area with different texture.Finally, a cascade classifier is used to discard redundant images and select key frames (frames with different features),so that a reduced capsule endoscopy video is obtained.The experimental results show that three performance indices of the algorithm,i.e.recall,precision and F-measure are 0.91,0.87 and 0.89 respectively,and the processing speed reaches 0.055 seconds per frame.And this algorithm achieves a better classification result and faster processing speed than that of the existing algorithms.