International Journal of Computer Applications (0975 – 8887) Volume 73– No.12, July 2013 22 Mosaic Evaluation: An Efficient and Robust Method based on Maximum Information Retrieval Kantilal P.Rane GF’s GCOE, Jalgaon (INDIA) Mayur B. Limbachiya GF’s GCOE, Jalgaon (INDIA) Shrirang S.Pandit GF’s GCOE, Jalgaon (INDIA) Sunil G. Bhirud VJTI, Mumbai (INDIA) ABSTRACT Performance Evaluation Process for video mosaic algorithms is developed on the basis of maximum information retrieval through closeness and residue between the original input images/ actual frames and the estimated images/frames from mosaic image. This evaluation method can be applicable to image as well as video mosaicing methods. Estimation of original input images/video frames and maximum information retrieval in terms of closeness/residue are the major steps involved in it. Without specific design of standard database, this method evaluates the mosaics in reference with the information in original input images/frames through a unique and single valued metric. Problems in case of mosaicing in complex condition like nonlinear vertical distortions and geometrical distortions in image and video capturing are discussed. Performance results are tested and compared with different mosaic images from different mosaic algorithms presented before. Keywords Mosaic evaluation; Image mosaic; Strip search. 1. INTRODUCTION Mosaic evaluations based on human judgment and personal analysis has been performed in early days. In latter days, image blurring [1] like ad hoc measures have been used but overall performance is not satisfactory. Uses of ground truth information of standard data sets were developed [2-6] during next days in various methods which included the various measuring metrics in 1 to 4-5 numbers. Estimation of standard data sets was a crucial at that time but very difficult task in these methods to cover real world data. Latter, very few methods [7-8] were introduced with no use of standard data set whose results are also satisfactory. Various parameter’s evaluation and various system’s ranking based on mosaic creation were proposed in literature like ranking of orientation tracking systems [9], ranking of electro- optical (EO) systems [10], Radiographic Quality [11], ranking of radiographic digital system [12], quality of size, shape and position of the image layer in radiographic panoramic images [13], quality of video compression [14] and ranking of tracking methods [15]. Based on the various methods proposed in literature, various requirements in evaluation methodologies are: 1) To find the single and unique valued metric with high speed and reliable operations so that ranking of mosaic method and quality evaluation of mosaics is feasible with real world data. 2) To develop the evaluation method for video mosaic to handle vertical distortions while propagating in horizontal direction. 3) To work without use of standard and reference databases for making the system computationally efficient. 4) To combine the evaluation of image mosaicing and video mosaicing. Mosaic evaluation should provide a measurable value for similarity between mosaic image and the input images/frames. It is required to determine some evaluative terms so that it will provide unique, accurate values for mosaic results. Problem is to develop mathematical model for evaluative terms. Model should be robust and should work on any image/video mosaicing technique. There is very less work reported on quantitative mosaic evaluation methodology without use of standard database. We tried to develop the evaluation method which is based on the information retrieval from the original images/frames. Depending on the information available in mosaic image, performance of image/video mosaic can be measured. Suitable steps of mosaic image evaluation includes, 1) Finding the maximum information areas (of original image/frame size) within the mosaic image, with respective to all the original image/frames of video which may be called Estimation of Video Frames/Input Images. 2) Combining the information present within all the estimated images/frames of video, this is suggested to be calculated in terms of percentage closeness and percentage residue. In case of image mosaicing, image transformation and image warping are generally wide, but it is harmful if images to be stitched are very large, which may lead to complex nature. Due to this one to one pixel based evaluation is even suitable in case of image mosaicing which indirectly can evaluate complexity. If the transformation is wide, complexity is more and should be indicated by the degradation of the ranking of mosaic method. So, same evaluation method can be applied to image as well as video mosaicing. Evaluation of nonlinear vertically distorted mosaic image from image stitching and video stitching may have different concept. In video mosaicing, overlapped portion is larger than that of in image mosaicing. So, with nonlinear vertical distortion, chances of direct correspondence of frames with mosaic in video mosaicing are less with respective to that of in image mosaicing. This indicates less evaluation performances in video mosaicing with respective to image mosaicing. Image mosaicing are generally used where geometrical distortions are more instead of video mosaicing. So, with geometrical distortions, evaluation performance of video mosaicing is exactly reverse (more) than that of in image mosaicing.