International Journal of Research in Engineering, Science and Management Volume-2, Issue-5, May-2019 www.ijresm.com | ISSN (Online): 2581-5792 289 Abstract: Image stitching is done by combining multiple input images such that the ideal result is a single image that contains contexts from all the inputs as well as seamless transitions between contexts. It is most commonly used in the field of panoramic photography. Large dissimilarities in the input image set are very problematic in image stitching applications. These cause several objections in the resulting output, such as ghosting and distortion. This paper presents a fully automated image stitching process that is aimed at reducing objections and increasing stitching quality. The proposed implementation adapted existing methods with an objective of having am implementation that is more robust against the dissimilarities caused by perspective, illumination, and occlusion. Keywords: image stitching, panorama. 1. Introduction Photographs are never exactly the same. Even in the exact same environment, in the right spot, with the right angle, and in the same lighting conditions, the image would contain, at least, minute differences throughout the scene. In reality, images of the same object would have varying perspectives, lighting conditions, and amount of occlusion. A panorama is an image that is created from several smaller images. This is done detecting and then making use of the similarities between images to join, or stitch, them in such a way that these images combine seamlessly. The study developed and tested an algorithm capable of combining two or more dissimilar images through existing stitching methods. For the purpose of this study, images are classified as dissimilar when they have slightly or extremely altered perspective, have different illumination or lighting conditions, and varying amount of occlusion. After images have been acquired, preprocessing of images is mandatory before they can be stitched. For example, the images can be projected onto a geometrical surface which can be on either a spherical, cylindrical, or planar surface. Camera-made distortions must also be corrected first before stitching can be done further [2]. The course of the image stitching process can be divided into two steps: image registration and image merging. During image registration, fraction of neighboring images are tested and compared to see if there are similar details between the images that can help in the alignment. After determining which parts of the images are most likely to be aligned, the images are lined up to form a panoramic image. A scenic image is constructed after images are successfully fused together. Hence, the three main processes are as follows: image acquisition, registration and fusion [2]. The study did not cover stitching of images which are not photographs or those images that are the products of computer graphics. It also did not cover the stitching of images which are contextually different. These are images that contain totally different subjects. Fig. 1. An example of image stitching. (top-left) and (top-right) are input images. (bottom) is the resulting image 2. Theoretical consideration There are many existing image stitching algorithms that can create panoramas by stitching similar images or images with higher number of similar keypoints. Most image stitching methods use the following succession of steps in the stitching process: detection of keypoints, matching keypoints, aligning images, and blending images. A. Keypoint detection David Lowe’s Scale Invariant Feature Transform (SIFT) is one approach in detecting the keypoints in an image. SIFT transforms image data into scale invariant and rotation invariant coordinates, and partially invariant to change in illumination and 3D camera viewpoint. In addition, SIFT provides robust detection even in the presence of affine distortion, resulting into distinctive key features [3]. In detecting the keypoints in an image, the stages that SIFT operator uses are as follows: local extrema detection, keypoint localization and orientation assignment [4]. One technique that can not only detect the keypoints but also match the detected keypoints is called Binary Robust Invariant Scalable Keypoints (BRISK). The method Image Stitching of Dissimilar Images P. Sannidhi 1 , Sathwik R. Gutti 2 , M. R. Shamanth 3 , R. Sai Charan 4 , S. K. Parikshith Nayak 5 1,2,3,4 Student, Department of Computer Science Engineering, Alva’s Inst. of Engg. and Tech., Moodbidri, India 5 Assistant Professor, Dept. of Computer Science Engineering, Alva’s Inst. of Engg. and Tech., Moodbidri, India