Abstract: System and method for joint refinement and perception of images are provided. A learning machine employs an image acquisition device for acquiring a set of training raw images. A processor determines a representation of a raw image, initializes a set of image representation parameters, defines a set of analysis parameters of an image analysis network configured to process the image's representation, and jointly trains the set of representation parameters and the set of analysis parameters to optimize a combined objective function. A module for transforming pixel-values of the raw image to produce a transformed image comprising pixels of variance-stabilized values, a module for successively performing processes of soft camera projection and image projection, and a module for inverse transforming the transformed pixels are disclosed. The image projection performs multi-level spatial convolution, pooling, subsampling, and interpolation.

Abstract: Image Signal Processing (ISP) optimization framework for computer vision applications is disclosed. The tuning of the ISP is performed automatically and presented as a nonlinear multi-objective optimization problem, followed by solving the problem using an evolutionary stochastic solver. An improved ISP of the embodiments of the invention includes at least features of search space reduction for reducing a number of ISP configurations, remapping the generated population to the reduced search space via mirroring, and global optimization function processing, which allow tuning all the blocks of the ISP at the same time instead of the prior art tuning of each ISP block separately. Also shown that an ISP tuned for image quality performs inferior compared with an ISP trained for a specific downstream image recognition task.

Abstract: A learning machine employs an image acquisition device for acquiring a set of training raw images. A processor determines a representation of a raw image, initializes a set of image representation parameters, defines a set of analysis parameters of an image analysis network configured to process the image's representation, and jointly trains the set of representation parameters and the set of analysis parameters to optimize a combined objective function. Processor executable instructions are organized into a module for transforming pixel-values of the raw image to produce a transformed image comprising pixels of variance-stabilized values, a module for successively performing processes of soft camera projection and image projection, and a module for inverse transforming the transformed pixels. The image projection process performs multi-level spatial convolution, pooling, subsampling, and interpolation.

Abstract: Method and system for denoising an edge in a raw digital image are described. First, a direction of a normal to the edge near a pixel p is determined, and neighboring pixels are projected to the normal to the edge, forming projected pixels. Then weighted intensities of the neighboring pixels are determined, including set of weights. One dimensional Gaussian filter centered on the pixel p and acting on the projected pixels is applied, and intensities of the neighboring pixels are adjusted according to the set of weights, resulting in a denoised value z of the pixel p using the weighted intensities, thereby denoising the edge of the raw image expediently.

Abstract: Methods and systems for denoising a digital image are provided. The method includes determining first and second plurality of pixel patches including respective first (p) and second (q) pixels, determining a patch distance between each pair of corresponding pixel patches in the first plurality and the second plurality of pixel patches, determining an effective distance between the first pixel (p) and the second pixel (q), repeating the above steps for the same first pixel (p) and another second pixel (q) until a predetermined number of pixels (q) in the raw digital image is processed, and then denoising the first pixel (p), including determining respective contributions of the pixels (q) into a noise reduction of the pixel (p), using respective effective distances of the pixels (q). A corresponding system is also provided. Embodiments of the invention provide computational advantages for denoising digital images.

Abstract: A method for determining a point spread function (PSF) of a camera, comprising: displaying on a display at least two images comprising visual features, elements having a different optical intensity, and a test pattern; acquiring the images using the camera; determining a coordinate transformation using the image comprising the visual features and its corresponding acquired image, the coordinate transformation for aligning together a coordinate system of the camera and a coordinate system of the display; aligning the acquired image comprising the elements having a different optical intensity using the determined coordinate system and determining a difference of intensity between the displayed image comprising the elements having a different optical intensity and the corresponding acquired image; aligning the acquired image comprising the test pattern and correcting an intensity of the acquired image comprising the test pattern, thereby obtaining a corrected image; and estimating a PSF using the corrected image