Detection Of Dysplasia From Endoscopic Images Using Daubechies 2 Wavelet Lifting Wavelet Transform

Abstract

We propose herein a new feature extraction method based on the lifting wavelet transform for dysplasia detection from an endoscopic image. In the proposed method, the input endoscopic image is converted into the hue-saturation-value color space, and the S space image is used. The pattern of the abnormal area is learned from this image using Daubechies 2 (db2) wavelet lifting wavelet transform. The lifting wavelet transform is performed on the detected image using the learned filter. Each frequency component is obtained using this method. The detected image generated from the sum of the high-frequency components is divided into small blocks. A static threshold is determined herein to obtain a binary image. Discrete wavelet transform is used to exclude smooth areas. V space images are used to exclude dark areas, such as shadows. This emphasizes the contour of the abnormal part. Finally, from the idea that the area surrounded by the outline is also abnormal, the life game is limitedly applied to emphasize the abnormal area. We describe the feature extraction in detail and present the experimental results demonstrating that our method is useful for the development of dysplasia detection from an endoscopic image.