Histogram variance controlled bleeding detectors for wireless capsule endoscopic images

Abstract

In this work, we propose histogram variance controlled bleeding detectors for detection of bleeding regions in wireless capsule endoscopic images. This new approach of bleeding detection is performed on CIE-L.a.b. color space model. This bleeding detection technique contains three modules. First module determines the histogram statistics of the L.a.b. component. Subsequently, variance controlled different Bleeding Detector functions (BDs) are proposed based on histogram statistical parameters. The final module describes K-means clustering algorithm which is used for the segmentation of bleeding regions. Real time endoscopic video datasets are used for this research. The results are validated by binary classifier with receiver operating curve (ROC) analysis. The experimental approach used in this research shows improvements and better results over the previous existing methods in both qualitative and quantitative evaluation (ROC) metrics.