Research
Research Interests
3D Television and Free-viewpoint TV was my major research area in my Ph.D., mainly focusing on efficient multiview coding and virtual view synthesis. Beyond 3DTV, I research image inpainting, image composition, image morphing, and image retrieval. Some of my major research threads are described below.
3DTV — Free-Viewpoint TV
Over the last couple of decades, 3DTV has gained a lot of popularity due to the quality of experience achieved through stereoscopy and autostereoscopy. The depth sensation in television is achieved by rendering two or more views simultaneously, which in some techniques also requires special glasses. It is desirable to use as few views as possible to achieve 3D vision, which poses new challenges for coding multi-views and generating virtual views from available views.
3D Video Coding
- Panorama view with spatiotemporal occlusion compensation for 3D video coding
- A panoramic 3D video coding with directional depth aided inpainting
3D View Synthesis & Visual Quality Enhancement
- Edge enhancement of depth based rendered image
- Depth Image Based Rendering with Inverse Mapping (VSIM)
- Edge shape enforcement for visual enhancement of depth image based rendering
3D Quality Assessment
- Synthesized Image Quality Evaluator (SIQE)
- Blind Depth Quality Metric (BDQM)
- DIBR-Synthesized Image Quality Metric (DSQM)
- Synthesized Image Quality Metric (SIQM)
Image Inpainting
Image inpainting is the removal of unwanted objects from digital images. The missing information in the unwanted region is filled using information from the neighborhood regions of the image. Inpainting is used in many research areas as a postprocessing step, in addition to recreational photography — for example, to fill holes in a novel view generated from a set of available views in Free-viewpoint TV.
Image Retrieval
Searching for an image, or searching for similar images to a given image, from a digital image repository is important in many applications, e.g. security, medicine, and defense. Content Based Image Retrieval (CBIR) uses color, texture, and shapes in the images to find the best matches.
- Content Based Image Retrieval Using Localized Multi-Texton Histogram — source code
Image De-Fencing
Detection and removal of fences from digital images becomes essential when an important part of the scene is occluded by such unwanted structures. Image de-fencing is challenging because manually marking fence boundaries is tedious and time-consuming. The fence is a distributed object and may cover a significant portion of the scene.
- Image De-Fencing Framework with Hybrid Inpainting Algorithm