Abstract: Systems and methods for background replacement in video using an occluded background prior that is dynamically maintained across frames. An input video frame is received at a separator, which reads previous background data from memory and generates a foreground matting (a per-pixel probability map indicating the foreground subject) based on the input frame and the previous background data. The system determines weights from the foreground matting and an accumulation map (a temporal exposure/confidence history), and updates the previous background data based on a weighted blending of the input frame and the previous background data. Updates for pixels classified as foreground are withheld to prevent leakage of the foreground subject into the background model. The updated background data is stored in memory for subsequent frames. Based on the foreground matting, a background replacer generates an output frame in which the foreground is preserved, and the background is replaced or modified.

Abstract: Systems and methods for determining a texture map for a high resolution image based on a downscaled low resolution version of the image. A texture map is used in image signal processing and image restoration systems to provide a location-specific indication of whether a particular area of an image includes high texture, low texture, edges, or flat regions. Using a low resolution version of the image, semantic cues are integrated with texture information to enable efficient, accurate, and cost-effective texture predictions and generate a texture classification map. The texture classification map based on the low resolution image is upscaled to a high resolution texture classification map. The texture classification map information can then be used to determine a selected filter to apply to each pixel and/or area in the high resolution image. The system provides spatially consistent decisions and improved noise robustness in texture to facilitate accurate image restoration.

Abstract: Techniques related to detecting texture content in an input image are discussed. Such techniques include accumulating pixel-wise directional gradient counts for positive and negative gradients for a first direction and positive and negative gradients for a second direction for pixels in a pixel window. A texture value for the target pixel of the window is determined based on a sum of a minimum of the positive and negative gradients for a first direction and a minimum of the positive and negative gradients for a second direction.

Abstract: Techniques related to detecting texture content in an input image are discussed. Such techniques include accumulating pixel-wise directional gradient counts for positive and negative gradients for a first direction and positive and negative gradients for a second direction for pixels in a pixel window. A texture value for the target pixel of the window is determined based on a sum of a minimum of the positive and negative gradients for a first direction and a minimum of the positive and negative gradients for a second direction.