Abstract: Generating an image with an imaging system includes capturing a first sub-image during a first exposure with a first pixel subset of an image sensor of the imaging system, capturing a second sub-image during a second exposure with a second pixel subset of the image sensor, capturing a third sub-image during a third exposure with a third pixel subset of the image sensor, and capturing a fourth sub-image during a fourth exposure with a fourth pixel subset of the image sensor. The first, second, third, and fourth exposures have different durations. A preferred exposure is selected from between the first, second, third and fourth exposures by analyzing the first, second, third, and fourth sub-images. Then, a full-resolution image is captured at the preferred exposure. An HDR image is generated based on the full-resolution image and the sub-images.

Abstract: A method is provided for image processing to remove noise and improve smoothing of nonimpulsive noise while preserving major identifying edges of the image, all of which are accomplished corresponding to a prescribed scale of interest. The method achieves nonlinear cluster filtering relying upon a maximum entropy principle. A single scale parameter governs filtering by dictating the spatial extent of neighboring data used for clustering. This, together with the signal characteristic (e.g., variation in gray scale) determines the scale parameter in the output space. The described method is unsupervised and data driven.