Improved Feature Point Pair Purification Algorithm Based on SIFT During Endoscope Image Stitching

Yan Liu; Jiawei Tian; Rongrong Hu; Bo Yang; Shan Liu; Lirong Yin; Wenfeng Zheng

doi:10.3389/fnbot.2022.840594

Improved Feature Point Pair Purification Algorithm Based on SIFT During Endoscope Image Stitching

Front Neurorobot. 2022 Feb 15:16:840594. doi: 10.3389/fnbot.2022.840594. eCollection 2022.

Authors

Yan Liu¹, Jiawei Tian¹, Rongrong Hu¹, Bo Yang¹, Shan Liu¹, Lirong Yin², Wenfeng Zheng¹

Affiliations

¹ School of Automation, University of Electronic Science and Technology of China, Chengdu, China.
² Department of Geography and Anthropology, Louisiana State University, Baton Rouge, LA, United States.

Abstract

Endoscopic imaging plays a very important role in the diagnosis and treatment of lesions. However, the imaging range of endoscopes is small, which may affect the doctors' judgment on the scope and details of lesions. Image mosaic technology can solve the problem well. In this paper, an improved feature-point pair purification algorithm based on SIFT (Scale invariant feature transform) is proposed. Firstly, the K-nearest neighbor-based feature point matching algorithm is used for rough matching. Then RANSAC (Random Sample Consensus) method is used for robustness tests to eliminate mismatched point pairs. The mismatching rate is greatly reduced by combining the two methods. Then, the image transformation matrix is estimated, and the image is determined. The seamless mosaic of endoscopic images is completed by matching the relationship. Finally, the proposed algorithm is verified by real endoscopic image and has a good effect.

Keywords: K-nearest; RANSAC; SIFT algorithm; endoscope; feature point matching; image mosaic.