Abstract:

Traditional distortion correction methods for medical endoscopic images have problems such as complication
of distortion modeling, high computation complexity, and susceptibility to errors. The proposed method uses a correction template. Sample coordinates are extracted from the template, and the distorted template image
acquired by the endoscope. These two kinds of extracted coordinates serve respectively as inputs and targets for
training to give the distortion model. Based on the distortion model, a one-to-one correspondence between pixels on the ideal and distorted images is set up. Image correction can then be accomplished by using interpolation. The experimental results show that the proposed method is simple, effective and accurate.

Key words: endoscopic image, nonlinear distortion, distortion correction, BP neural network (BPNN)