Investigating and Analyzing Refinements to Image Completion Processes
Image completion techniques restore the damaged portion in an image especially for large images. Our goal is to implement and further refine an image completion algorithm that can be efficiently used for providing privacy to participants in social networking sites. To achieve our goal, we need to identify a suitable approach and successfully implement it. Moreover, we need to determine a methodology to investigate its relevance to dealing with large images. In addition, we must investigate improvements to the efficiency of the inpainting operation. The last, we need to determine the quality of the repaired images is unaffected by the optimizations that we introduced.
We have successfully implemented Criminisi’s exemplar-based image completion method. We have developed an image bank to test the system, which shows that the patch matching is a time consuming process with time complexity O (n^2). In order to accelerate the patch matching scheme, we introduce a new approximate patch matching process based on a set of human defined hash functions. Our approach reduces at least 50% time of Criminisi’s approach testing on our images bank, which achieve O (n). Our approach can achieve a similar or better result than Criminisi’s algorithm by using the measure SNR.
 Shaun Bangay and Orson Lv. Evaluating locality sensitive hashing for matching partial image patches in a social media setting. Journal of Multimedia, 9(1):14–24, 2014. [PDF] [BibTeX]
 Zongming Lv. Investigating and analyzing refinements of image completion processes. Technical Report Honours Project Report, GIVE group, School of Information Technology, Deakin University, Australia, April 2012. [PDF] [BibTeX]