This article was automatically translated from the original Turkish version.
Image mosaic stitching is the process of combining multiple images of the same scene into a single image with greater detail, by aligning them according to overlapping common regions. To capture a single high-resolution frame of a large scene, a camera with high resolution must be positioned at a sufficient distance. However, in enclosed spaces, the available distance for camera placement is limited, and high-resolution cameras are expensive; therefore, image mosaic stitching is employed to capture large areas. This technique enables the creation of high-resolution images by combining multiple images captured with standard-resolution cameras.
Image mosaic stitching can be used for various purposes, including obtaining higher-quality images at lower cost, photographing areas too large to fit within a single camera opinion field of view, creating panoramic images, and removing moving objects from a video sequence. The main application areas of image mosaic stitching are listed below:
Depending on the number of images to be combined, the image mosaic stitching process typically occurs step-by-step by aligning two images at a time. One image is designated as the reference image, and the other as the target image. The target image is transformed to align matching points with those in the reference image. The transformed image is then appended to the reference image to form a larger composite image. This resulting mosaic image becomes the new reference image in the next step, enabling multi-image mosaic stitching.
Figure 1. Image mosaic stitching process (Riedl, Zipper, Meier and Diedrich, 2014)
Different methods have been developed to perform image mosaic stitching, broadly categorized into spatial domain and frequency domain approaches. Spatial domain methods can be further divided into area-based and feature-based methods. Area-based methods match regions between two images based on intensity values; however, they are significantly affected by changes in scale, lighting, and position. The SIFT method, within the category of feature-based methods, overcomes these limitations.
Figure 2. Steps of feature-based image mosaic stitching (Guan, Yang, Chen, Dai and Wang, 2016)
In feature-based image matching, distinctive keypoints are first detected in each image, and corresponding feature descriptors are extracted. Points with similar descriptors are then matched across images, establishing a relationship between them. Since some incorrect matches may still occur, outliers are removed based on consistency criteria. Using the remaining reliable matches, a homography transformation is estimated. This matrix is then applied to warp and align the images. Finally, color blending is performed to correct color inconsistencies between the images.
