2011/09/21 by Alper Ayvacı, Stefano Soatto, Ayvaci, Alper +1 · 1 citation
Computer Science · #Advanced Image and Video Retrieval Techniques #Visual Attention and Saliency Detection #Advanced Neural Network Applications
paper · pdf · doi:10.48550/arxiv.1109.4683
We describe an approach for segmenting an image into regions that correspond\nto surfaces in the scene that are partially surrounded by the medium. It\nintegrates both appearance and motion statistics into a cost functional, that\nis seeded with occluded regions and minimized efficiently by solving a linear\nprogramming problem. Where a short observation time is insufficient to\ndetermine whether the object is detachable, the results of the minimization can\nbe used to seed a more costly optimization based on a longer sequence of video\ndata. The result is an entirely unsupervised scheme to detect and segment an\narbitrary and unknown number of objects. We test our scheme to highlight the\npotential, as well as limitations, of our approach.\n