High Performance Feature Transformation Architecture based on Bag-of-Features in CAD system for Colorectal Endoscopic Images

Our research target to a computer-aided diagnosis (CAD) system for colorectal endoscopic images with narrow band imaging (NBI) magnification, which identifies a pathology type from local feature in the NBI endoscopic image. We propose a high speed feature transformation for CAD system by using Manhattan distance calculation and on the fly normalization method. A high performance and low cost algorithm for multiple Scan Window (SW) processing for FPGA is also introduced. The proposed high speed feature transformation can complete the transformation processing within 380 msec on a real time Full HD NBI endoscopic image.