Epitome and its applications

probabilistic graphical models;image synthesis;recognition;colorization

Chu, Xinqi ; Huang ; Thomas S.

关键词: probabilistic graphical models;    image synthesis;    recognition;    colorization;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/42279/Xinqi_Chu.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Due to the lack of explicit spatial consideration, the existing epitome modelmay fail for image recognition and target detection, which directly motivates us to propose the so-called spatialized epitome in this thesis. Extendedfrom the original simple graphical model of epitome, the spatialized epitome provides a general framework to integrate both appearance and spatialarrangement of patches in the image to achieve a more precise likelihood representation for image(s) and eliminate ambiguities in image reconstructionand recognition. From the extended graphical model of epitome, a new EMlearning procedure is derived under the framework of variational approximation. The learning procedure can generate an optimized summary of theimage appearance based on patches and automatically cluster the spatialdistribution of the similar patches. From the spatialized epitome, we presenta principled (parameter-free) way of inferring the probability of a new inputimage under the learned model and thereby enabling image recognition andtarget detection. We show how the incorporation of spatial information enhances the epitome’s ability for discrimination on several tough vision tasks,e.g., misalignment/cross-pose face recognition, and vehicle detection with afew training samples. We also apply this model to image colorization whichnot only increases the visual appeal of grayscale images, but also enrichesthe information contained in scientific images that lack color information.Most existing methods of colorization require laborious user interaction forscribbles or image segmentation. To eliminate the need for human labor, wedevelop an automatic image colorization method using epitome. Built upona generative graphical model, epitome is a condensed image appearance andshape model which also proves to be an effective summary of color information for the colorization task. We train the epitome from the reference imagesand perform inference in the epitome to colorize grayscale images, renderingbetter colorization results than previous methods.