An intelligent model for visual scene analysis and compression

This paper presents an improved approach for indicating visually salient regions of an image based upon a known visual search task. The proposed approach employs a robust model of instantaneous visual attention (i.e. “bottom-up”) combined with a pixel probability map derived from the automatic detection of a previously-seen object (task-dependent i.e. (“top-down”). The objects to be recognized are parameterized quickly in advance by a viewpoint-invariant spatial distribution of Speeded Up Robust Features (SURF) interest-points. The bottom-up and top-down object probability images are fused to produce a task-dependent saliency map. The proposed approach is validated using observer eye-tracker data collected under object search-and-count tasking. Proposed approach shows 13% higher overlap with true attention areas under task compared to bottom-up saliency alone. The new combined saliency map is further used to develop a new intelligent compression technique which is an extension of Discrete Cosine Transform (DCT) encoding. The proposed approach is demonstrated on surveillance-style footage throughout.