EDGE-AWARE SPATIO-TEMPORAL FILTERING AND OPTICAL FLOW ESTIMATION IN REAL TIME

The disclosure provides an approach for edge-aware spatio-temporal filtering. In one embodiment, a filtering application receives as input a guiding video sequence and video sequence(s) from additional channel(s). The filtering application estimates a sparse optical flow from the guiding video sequence using a novel binary feature descriptor integrated into the Coarse-to-fine PatchMatch method to compute a quasi-dense nearest neighbor field. The filtering application then performs spatial edge-aware filtering of the sparse optical flow (to obtain a dense flow) and the additional channel(s), using an efficient evaluation of the permeability filter with only two scan-line passes per iteration. Further, the filtering application performs temporal filtering of the optical flow using an infinite impulse response filter that only requires one filter state updated based on new guiding video sequence video frames. The resulting optical flow may then be used in temporal edge-aware filtering of the additional channel(s) using the nonlinear infinite impulse response filter.

BACKGROUNDField of the Invention

Embodiments of the disclosure presented herein relate to image and video processing and, more specifically, to efficient edge-aware spatio-temporal filtering and optical flow estimation that may be performed in real time.Description of the Related Art

Edge-aware filters are an important tool in many image and video applications, such as high-dynamic range (HDR) tone mapping, stylization, and detail manipulation. As used herein, filtering generally refers to combining pixel values, over space or time, according to a filter kernel that defines filter weights. Edge-aware filters in particular make use of information about image edges in the filtering process and are space-variant (i.e., their kernel is dependent on the input data).

Traditional spatio-temporal edge-aware filters have two major drawbacks that impede their use in real-time filtering of videos. First, such filters either operate iteratively on complete video volumes (i.e., all the frames in a video) or on sliding windows of video frames, which require the entire video volume or sliding window of video frames to be stored in memory, thereby incurring high processing latency and requiring access to a large memory with high bandwidth. Second, such filters require accurate optical flow information as input in order to align neighboring frames for temporal filtering, but accurate optical flow is difficult to obtain efficiently with traditional techniques.