Abstract: In various embodiments, an image processing method is provided. In those embodiments, a first image and a second image are obtained. A conjugate image is obtained based on the second image. In the conjugate image, a given pixel has a luminance less than a corresponding pixel in the second image. In those embodiments, a weight is obtained based on the conjugate image. During an image fusion of the first and the second image, the weight is applied - for example for assigning percentages of luminance coming from the first and second images. In those embodiments, as a result of the weight, for certain parts like vegetation, the fused image is biased towards the color image, while for some other parts, the fused image is biased towards the NIR image. Other embodiments may include computer systems, apparatus, and computer programs stored storage medium, each configured to perform the image processing method.

Abstract: This application is directed to fusion of two images (e.g., a near infrared (NIR) image and an RGB image) that are captured simultaneously in a scene. A computer system extracts a first luminance component and a first color component from the first image, and extracts a second luminance component from the second image. An infrared emission strength is determined based on the first and second luminance components. The computer system combines the first and second luminance components based on the infrared emission strength to obtain a combined luminance component. The combined luminance component is combined with the first color component to obtain a fused image.

Abstract: In some embodiments, an image processing application receives a first image and a second image for fusing. The image processing application generates a foreground object mask for a foreground object in the first image and dilates the foreground object mask to generate a dilated mask. The image processing application further integrates the image content of the first image in at least the dilated mask with the second image to generate an updated second image.

Abstract: A computer-implemented method of training a neural network comprises: identifying, by one or more processors, a set of parameters for a cost function for the neural network; applying, by the one or more processors, a gradient optimization of the cost function to generate a set of intermediate optimized parameter values for the cost function; initializing, by the one or more processors, a simulated annealing optimization of the cost function using the set of intermediate optimized parameter values for the cost function; applying, by the one or more processors, the simulated annealing optimization of the cost function to generate a set of final optimized parameter values for the cost function; assigning, by the one or more processors, the set of final optimized parameter values as weights for connections of the neural network; and using, by the one or more processors, the neural network to perform operations.

Abstract: A computer-implemented method of training a neural network comprises: identifying, by one or more processors, a set of parameters for a cost function for the neural network; applying, by the one or more processors, a gradient optimization of the cost function to generate a set of intermediate optimized parameter values for the cost function; initializing, by the one or more processors, a simulated annealing optimization of the cost function using the set of intermediate optimized parameter values for the cost function; applying, by the one or more processors, the simulated annealing optimization of the cost function to generate a set of final optimized parameter values for the cost function; assigning, by the one or more processors, the set of final optimized parameter values as weights for connections of the neural network; and using, by the one or more processors, the neural network to perform operations.