Sub -pixel registration and image analysis with application to image comparison
Abstract
It has been known that the image analysis can provide efficient methodology for various fields. Among various application of image analysis, image comparison is one of the important areas that can be applied. Image comparison is required to find the similarities or differences between image pairs. The classical way for comparison is the visual inspection by human observers. But visual comparison is subjective and impractical when the number of image pairs are huge. In this case, autonomous image comparator is preferable. Also, image analysis can offer systemic way to find the one-to-one correspondence of each pixels. In this thesis, we have studied three different application of image analysis. ^ In the first part of this research, we propose sub-pixel registration algorithm. Our target application is to register two images taken by different quality imaging device such as digital camera. Different imaging devices produce shift, rotation, scaling and other nonlinear distortions although they contain the same scene. To overcome these distortion and to register them, we used region splitting and histogram matching method. Proposed algorithm showed robust performance for both digital camera images and scanner images. ^ In the second and third part, we propose autonomous image comparator. The images that need to be compared can be hard-copy or soft-copy depending on the application. The hard-copy form is the result of image reproduction device like printers and we want to see how good is the reproduced image. For this objective, we need to convert hard-copy image into soft-copy using another imaging device such as scanner. In the second part of this thesis, we propose image comparison algorithm under these circumstances. In addition, we proposed several preprocessing algorithm to reduce the effects of the scanner and mistakes made by end users. Our proposed algorithm is also another application of objective measurement for the goodness of the reproduced image. ^ We also studied for the soft-copy image comparison. In this case, the reproduced images are unpredictably distorted. The distortion can be noticeable or unnoticeable and may depend on the contents of the image. Our goal is to prescreen the noticeable differences. To distinguish noticeably different image pairs, we proposed segmentation based image comparison algorithm. We proposed segmentation algorithm that is suitable for our comparison image pairs. The comparison is made based on the segmentation map. With these procedures, the comparator can distinguish noticeable and unnoticeable differences. With the proposed algorithm, we do not need to inspect each image visually. Many image pairs can be replaced by computer algorithm. This tool is efficient for the automation of image comparison when the number of image pairs are huge.^
Degree
Ph.D.
Advisors
Jan Allebach, Purdue University.
Subject Area
Engineering, Electronics and Electrical
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.