Abstract: Image pair co-registration systems and methods include receiving a pair of multi-modal images, defining a parametric deformation model, defining a loss function that is minimized when the pair of images are aligned, and performing a multi-scale search to determine deformation parameters that minimize the loss function. The optimized deformation parameters define an alignment of the pair of images. The pair of images may include visible spectrum image and an infrared image. The method further includes resizing the visible spectrum image to match the infrared image, applying at least one lens distortion correction model, and normalizing a dynamic range of each of the pair of images. The multi-scale search may further include resizing the pair of images to a current processing scale, applying adaptive histogram equalization to the pair of images to generate equalized images, applying Gaussian Blur to the equalized images, and optimizing the deformation parameters.

Abstract: Techniques are disclosed for panoramic image construction based on images captured by rotating imagers. In one example, a method includes receiving a first sequence of images associated with a scene and captured during continuous rotation of an image sensor. Each image of the first sequence has a portion that overlaps with another image of the first sequence. The method further includes generating a first panoramic image. The generating includes processing a second sequence of images based on a point-spread function to mitigate blur associated with the continuous rotation to obtain a deblurred sequence of images, and processing the deblurred sequence based on a noise power spectral density to obtain a denoised sequence of images. The point-spread function is associated with the image sensor's rotation speed. The second sequence is based on the first sequence. The first panoramic image is based on the denoised sequence.

Abstract: Systems and methods for registration of image pairs, including camera response function normalization are disclosed and include selecting a pair of images from a set of images, each of the pair of images having associated metadata, determining a camera response function for each of the images in the pair of images using the associated metadata, normalizing each camera response function for across the set of images, and applying the normalized camera response function to the pair of images. A deformation map is generated in a multi-scale process using registration parameters, to deform one of the pair of images to align with another of the pair of images. The image pairs may be selected by identifying image pairs having an overlap that exceeds an overlap threshold, having a sequential proximity in an image series satisfying a proximity threshold, and/or having estimated image capture attitudes within an attitude threshold.

Abstract: Various techniques are disclosed for separating and removing low-frequency shadow or shading (also referred to herein as “non-uniformity”) from images that have been corrupted by the non-uniformity. A non-uniformity estimate that approximates the non-uniformity effect on the corrupted image may be generated by iteratively adding new blotches of non-uniformity data represented by two-dimensional (2D) functions, such as 2D Gaussian functions, to the non-uniformity estimate and applying filters to smoothen the 2D functions. In each iteration of the non-uniformity estimate generation process, a new non-uniformity update candidate that minimizes a cost function is identified. The corrupted image is processed based on the non-uniformity estimate to generate a corrected image.

Abstract: Techniques are disclosed for panoramic image construction based on images captured by rotating imagers. In one example, a method includes receiving a first sequence of images associated with a scene and captured during continuous rotation of an image sensor. Each image of the first sequence has a portion that overlaps with another image of the first sequence. The method further includes generating a first panoramic image. The generating includes processing a second sequence of images based on a point-spread function to mitigate blur associated with the continuous rotation to obtain a deblurred sequence of images, and processing the deblurred sequence based on a noise power spectral density to obtain a denoised sequence of images. The point-spread function is associated with the image sensor's rotation speed. The second sequence is based on the first sequence. The first panoramic image is based on the denoised sequence.

Abstract: Various techniques are disclosed for systems and methods to provide image resolution enhancement. For example, a method includes: receiving a reference image (e.g., a visible light image) of a scene comprising image pixels identified by pixel coordinates; receiving a lower-resolution target image (e.g., an infrared image) of the scene; resizing the target image to a larger size; determining an adaptive-shape neighborhood for each pixel coordinate, wherein the adaptive-shape neighborhood extends from the each pixel coordinate such that those reference image pixels that are within the shape-adaptive neighborhood meet a regularity condition; determining, for each adaptive-shape neighborhood, a local estimate based on those target image pixels that are within the adaptive-shape neighborhood; and aggregating the local estimates associated with the adaptive-shape neighborhoods to provide a global estimate that corresponds to the target image with an improved resolution.

Abstract: Various techniques are disclosed for separating and removing low-frequency shadow or shading (also referred to herein as “non-uniformity”) from images that have been corrupted by the non-uniformity. A non-uniformity estimate that approximates the non-uniformity effect on the corrupted image may be generated by iteratively adding new blotches of non-uniformity data represented by two-dimensional (2D) functions, such as 2D Gaussian functions, to the non-uniformity estimate and applying filters to smoothen the 2D functions. In each iteration of the non-uniformity estimate generation process, a new non-uniformity update candidate that minimizes a cost function is identified. The corrupted image is processed based on the non-uniformity estimate to generate a corrected image.

Abstract: Various techniques are disclosed for systems and methods to provide image resolution enhancement. For example, a method includes: receiving a reference image (e.g., a visible light image) of a scene comprising image pixels identified by pixel coordinates; receiving a lower-resolution target image (e.g., an infrared image) of the scene; resizing the target image to a larger size; determining an adaptive-shape neighborhood for each pixel coordinate, wherein the adaptive-shape neighborhood extends from the each pixel coordinate such that those reference image pixels that are within the shape-adaptive neighborhood meet a regularity condition; determining, for each adaptive-shape neighborhood, a local estimate based on those target image pixels that are within the adaptive-shape neighborhood; and aggregating the local estimates associated with the adaptive-shape neighborhoods to provide a global estimate that corresponds to the target image with an improved resolution.

Abstract: Various techniques are disclosed to suppress distortion in images (e.g., video or still images), such as distortion caused by atmospheric turbulence. For example, similar image blocks from a sequence of images may be identified and tracked along motion trajectories to construct spatiotemporal volumes. The motion trajectories are smoothed to estimate the true positions of the image blocks without random displacements/shifts due to the distortion, and the smoothed trajectories are used to aggregate the image blocks in their new estimated positions to reconstruct the sequence of images with the random displacements/shifts suppressed. Blurring that may remain within each image block of the spatiotemporal volumes may be suppressed by modifying the spatiotemporal volumes in a collaborative fashion.

Abstract: Various techniques are disclosed to suppress distortion in images (e.g., video or still images), such as distortion caused by atmospheric turbulence. For example, similar image blocks from a sequence of images may be identified and tracked along motion trajectories to construct spatiotemporal volumes. The motion trajectories are smoothed to estimate the true positions of the image blocks without random displacements/shifts due to the distortion, and the smoothed trajectories are used to aggregate the image blocks in their new estimated positions to reconstruct the sequence of images with the random displacements/shifts suppressed. Blurring that may remain within each image block of the spatiotemporal volumes may be suppressed by modifying the spatiotemporal volumes in a collaborative fashion.