We propose a novel tool for re-rendering objects in indoor scene images with new textures. It aims to address the problem of too much manual work of positioning and alignment when applying new texture onto an object surface in an indoor scene image. The algorithm of the tool is based on establishing 2D projective transformation between texture images and planar object surfaces in scene images. In order to find the transformation, we use a sampled rectangular texture pattern from a large synthesized planar texture and a planar quadrangle corresponding to object surface orientation estimation, which is generated by a geometric orientation hypothesis framework. The tool also puts effort in adjusting the scaling and reducing artifacts for re-rendered textures. We present the re-rendering results for ceilings, walls, floors, etc. that naturally correspond to room geometry layout.
Category search is a searching activity where the user has an example image and searches for other images of the same category. This activity often requires appropriate keywords of target categories making it difficult to search images without prior knowledge of appropriate keywords. Text annotations attached to images are a valuable resource for helping users to find appropriate keywords for the target categories. We propose an image exploration system in this article for category image search without the prior knowledge of category keywords. Our system integrates content-based and keyword-based image exploration and seamlessly switches exploration types according to user interests. The system enables users to learn target categories both in image and keyword representation through exploration activities. Our user study demonstrated the effectiveness of image exploration using our system, especially for the search of images with unfamiliar category compared to the single-modality image search. © 2016 Society for Imaging Science and Technology.