Technical Assessment of Various Image Stitching Techniques: A Deep Learning Approach

Abstract

This chapter presents various image-stitching techniques based on different feature extraction methods. Image stitching can be regarded as a process of assembling more than one image of the same scene having an overlapping area in between them into a single high-resolution image. Direct as well as feature-based methods, are two broad categories of image stitching 52techniques. Direct methods are more time consuming as compared to feature-based techniques since it needs to compare each and every pixel intensities with each other. Image stitching on the overlapping region is divided into the following steps: first features are detected and described using any kind of feature extraction; secondly, find matching pairs followed by removing mismatches by RANSAC (random sample consensus), then estimate the homography matrix; and finally, blend the overlapping areas. The authors introduce a complete framework for feature-based automatic image stitching. Stitching images where overlapping regions are missing is also a very hot topic of research. It follows image extrapolation, alignment, and in painting.