基于滤波器组与中心对称局部二值模式的内窥镜病灶图像检测方法

Focus image detection in endoscopy video based on filter group and center symmetric local binary pattern

为了解决在大量内窥镜图像中准确检测多种病灶这一难题,提出一种滤波器组与中心对称局部二值模式的多重纹理检测方法。首先将内窥镜图像经过LM(Leung-Malik)滤波器组,计算其响应的联合概率分布作为图像的纹理特征;采用CS-LBP对该图像进行编码,从另一角度提取其纹理特征。将训练样本的两种纹理特征结合进行聚类,形成一组纹理基元构建纹理基元字典,通过计算像素与纹理基元的欧氏距离获得基于纹理基元的图像直方图。最后,用K-邻近算法对纹理图像进行分类。实验结果表明,该方法特异度和灵敏度分别达到88%和92%,优于已有结果,可代替应用于内窥镜图像的临床初步检测。

In order to solve the problem of accurate detection of multiple focuses in a large number of endoscopic images,a multiple texture detection method based on filter group and center symmetric local binary pattern( CS-LBP) is proposed. Through the LeungMalik( LM) filter group,the endoscope image is calculated,the response of the joint probability distribution is made as texture features of the image. Using the image encoding CS-LBP,the texture features are extracted from another angle. The clustering combined with two texture features of training samples,which forms a set of texture element text on dictionary,and the image texture histogram based on element is obtained by calculating the pixel and the texture elements Euclidean distance. Finally,the KNN( K-nearest neighbor)algorithm is used to classify the texture image. The experimental results show that the specificity and sensitivity of the proposed method is88% and 92% respectively,which is superior to the existing results. It can be used for clinical preliminary detection of endoscope image.