Hi-speed computational method for removing sea-fog on the images using graphic processors and parallel processing

摘要

The present invention relates to a fog removal acceleration method based on parallel computation using a graphic processor. The fog removal acceleration method is used to remove fog from input images in a system including a device with parallel processing capabilities and a host processor. The fog removal acceleration method includes the following steps: (a) extracting dark channels from the input images; (b) setting the values of pixels having a brightness within a predetermined range with respect to the brightest spot in the extracted dark channels as atmospheric scattered light; (c) normalizing the images by dividing the values of individual color values of the input images by the atmospheric scattered light values and extracting dark channels from the normalized images again; (d) obtaining the transmission amount by subtracting the values of the re-extracted dark channels from predetermined constant values while multiplying a predetermined weight value to the transmission amount in the subtraction operation; (e) applying a bi-directional filter to refine the transmission amount; and (f) removing fog by using the refined transmission amount to restore the input images. The steps (a), (d), and (f) are executed by the host processor while the step (e) is executed by the graphic processor. The processes of refining the transmission amount in the computations for fog removal can be processed in a parallel manner through the graphic processor, thereby enabling quicker computation.