Effects of Noise and Relative Overlap on Image Mosaicing using SURF Features

Performance of image mosaicing depends on overlap between the images to be joined and the percentage of noise present. An algorithm is used for joining images based on image matching by comparing the descriptors for different images. This paper is concerned with the analysis of effect of variations in noise and degree of relative overlap on the algorithm and obtaining their limits. Speeded Up Robust Features (SURF) have been used for key point detection. Percentage overlap between the images to be joined is varied to find out the minimum value required for mosaicing. Further, relative overlap is kept constant and noise is increasingly added to the input images to find out the maximum amount of noise the algorithm can sustain. The aforementioned experiments were performed over a set of images. Ultimately, a range for maximum permissible noise and minimum overlap required is defined for acceptable panorama generation.