ADAPTIVE METHOD FOR SPACE-TIME FILTERING Anisotropic Bilateral VIDEO

摘要

1. method of adaptive space-time u0430u043du0438u0437u043eu0442u0440u043eu043fu043du043eu0439 bilateral filtering of video signals, in which the u0434u043bu00a0 each (current) u043fu0438u043au0441u0435u043bu00a0 u0438u0437u043eu0431u0440u0430u0436u0435u043du0438u00a0 u0432u044bu043fu043eu043bu043du00a0u044e t the following operations:;in the current u043fu0438u043au0441u0435u043bu00a0 around the first area u0441u043eu0441u0442u043eu00a0u0449u0443u044e from current u043fu0438u043au0441u0435u043bu00a0 and several nearby neighboring pixels u0438u0437u043eu0431u0440u0430u0436u0435u043du0438u00a0;;choose two lines, one of which is horizontally and vertically (u0433u043eu0440u0438u0437u043eu043du0442u0430u043bu044cu043du0430u00a0 u043bu0438u043du0438u00a0) u0434u0440u0443u0433u0430u00a0 (u0432u0435u0440u0442u0438u043au0430u043bu044cu043du0430u00a0 u043bu0438u043du0438u00a0) through the current pixel and the pixel area and the current u043fu0438u043au0441u0435u043bu00a0;;u043eu043fu0440u0435u0434u0435u043bu00a0u044eu0442 horizontal gradient along the horizontal lines of the current u043fu0438u043au0441u0435u043bu00a0 as the difference between the u043fu0438u043au0441u0435u043bu0435u043c located to the left of the current u043fu0438u043au0441u0435u043bu00a0 and p u0438u043au0441u0435u043bu0435u043c located to the right of the current u043fu0438u043au0441u0435u043bu00a0;;u043eu043fu0440u0435u0434u0435u043bu00a0u044eu0442 vertical gradient along the vertical line of the current u043fu0438u043au0441u0435u043bu00a0 as the difference between the u043fu0438u043au0441u0435u043bu0435u043c located on top of the current u043fu0438u043au0441u0435u043bu00a0 and peaks u0435u043bu0435u043c located below the current u043fu0438u043au0441u0435u043bu00a0;;u043eu043fu0440u0435u0434u0435u043bu00a0u044eu0442 u043au043eu0440u0440u0435u043bu00a0u0446u0438u043eu043du043du0443u044e matrix u0434u043bu00a0 horizontal gradient and vertical gradients of the current u043fu0438u043au0441u0435u043bu00a0, and also own the vectors l, s and u0441u043eu0431u0441u0442u0432u0435u043du043du044b e values (l, s u043au043eu0440u0440u0435u043bu00a0u0446u0438u043eu043du043du043eu0439 matrix;;u043eu043fu0440u0435u0434u0435u043bu00a0u044eu0442 force edge contours _ power formula;where n is the number of the case sensitivity to the contour of the object in the image.;the contour of the object in the image u043eu043fu0440u0435u0434u0435u043bu00a0u044eu0442 direction as the angle between its own vector s u043au043eu0440u0440u0435u043bu00a0u0446u0438u043eu043du043du043eu0439 matrix with smaller amounts of their own values and mr. u043eu0440u0438u0437u043eu043du0442u0430u043bu044cu043du043eu0439 line;;u0434u043bu00a0 each frame of the number of neighboring frames, u0432u043au043bu044eu0447u0430u00a0 current frame, choose around u043fu0438u043au0441u0435u043bu00a0, the current u043fu0438u043au0441u0435u043bu044e, search area, u0441u043eu0441u0442u043eu00a0u0449u0443 yu from the current u043fu0438u043au0441u0435u043bu00a0 and nearest neighbouring pixels u0438u0437u043eu0431u0440u0430u0436u0435u043du0438u00a0;;u0432u044bu0447u0438u0441u043bu00a0u044eu0442 weighting factors of the filter by u0441u0440u0430u0432u043du0435u043du0438u00a0 first area (window size of the kk) in the current frame with the same size in the field u043eu0431u043bu0430u0441u0442u00a0u043cu0438 search the window size of the nn) in the current and neighboring personnel;;counting rates of adaptation of temporal and u0431u0438u043bu0430u0442u0435u0440u0430u043bu044cu043du043eu0433u043e filters for power circuits, and the adaptation of t, the temporary filter to the power circuit at u044bu0447u0438u0441u043bu00a0u044eu0442 formula, etc. u00a0u0432u043bu00a0u0435u0442u0441u00a0 adjustable parameter and rate adaptation (r u0431u0438u043bu0430u0442u0435u0440u0430u043bu044cu043du043eu0433u043e filter to the energy contour of u0432u044bu0447u0438u0441u043bu00a0u044eu0442 formula, where kb is r u0435u0433u0443u043bu0438u0440u0443u0435u043cu044bu0439 parameter;;u0432u044bu043fu043eu043bu043du00a0u044eu0442 filtering in time by u043du0430u0445u043eu0436u0434u0435u043du0438u00a0 weighted sum of pixels, which u043du0430u0445u043eu0434u00a0u0442u0441u00a0 for prospecting in the current and neighboring hr, and the result u0444u0438u043bu044cu0442u0440u0430u0446 and u0441u043eu0445u0440u0430u043du00a0u044eu0442 in variable out _ pixelT;;u043eu043fu0440u0435u0434u0435u043bu00a0u044eu0442 second area as the area of the u043fu0438u043au0441u0435u043bu00a0, in which the current u043fu0438u043au0441u0435u043bu044e value out _ pixelT;;u0440u0430u0437u0434u0435u043bu00a0u044eu0442 second area in the two u043du0430u043fu0440u0430u0432u043bu0435u043du0438u00a0 subarea along first and second subarea.;u0432u044bu043fu043eu043bu043du00a0u044eu0442 bilateral filtering u0434u043bu00a0 first and second subarea according to the following formula;where is the intensity of the point with index i,;ip is the intensity of the point u0434u043bu00a0 which u0432u044bu043fu043eu043bu043du00a0u0435u0442u0441u00a0 u0444u0438u043bu044cu0442u0440u0430u0446u0438u00a0, as well as h (spatial) use u00a0u0434u0440u043e low-frequency filter;;u0432u044bu0447u0438u0441u043bu00a0u044eu0442 bilateral filtering result as weighted summation results of filtering the first and second subarea.;u0432u044bu043fu043eu043bu043du00a0u044eu0442 approximation of the u043fu0438u043au0441u0435u043bu00a0 u043fu0438u043au0441u0435u043bu00a0u043cu0438 located along the u043du0430u043fu0440u0430u0432u043bu0435u043du0438u00a0 contour, with an average u0430u043fu043fu0440u043eu043au0441u0438u043cu0430u0446u0438u00a0 u043fu0440u0435u0434u0441u0442u0430u0432u043bu00a0u0435u0442 adjacent to the current u043fu0438u043au0441u0435u043bu044e pixels located along the contour.;u0432u044bu0447u0438u0441u043bu00a0u044eu0442 final result as the weighting sum of bilateral filtering and approximation along the contour.;2. method for u043eu0442u043bu0438u0447u0430u044eu0449u0438u0439u0441u00a0 1, so that the u0434u043bu00a0 u0441u0440u0430u0432u043du0435u043du0438u00a0 areas of pixels using the measure;where inp1 (xn, yn, tn) refers to the pixel of the current frame xn tn coordinates and yn,;inp2 (p, s, tn + (r) denotes the pixel from the frame tn + r coordinates of r and s, the index (i u0438u0437u043cu0435u043du00a0u0435u0442u0441u00a0 ranging from to in the horizontal direction.;the index j u0438u0437u043cu0435u043du00a0u0435u0442u0441u00a0 ranging from to in the vertical direction.;3. method for u043eu0442u043bu0438u0447u0430u044eu0449u0438u0439u0441u00a0 1, so that a filter u0432u044bu043fu043eu043bu043du00a0u044eu0442 as follows;,;where denotes the pixel from the image coordinates xn tn + r + r and yn + s;the index r u0438u0437u043cu0435u043du00a0u0435u0442u0441u00a0 ranging from to in the horizontal direction.;the index's u0438u0437u043cu0435u043du00a0u0435u0442u0441u00a0 ranging from to in the vertical direction.;the index r u0438u0437u043cu0435u043du00a0u0435u0442u0441u00a0 ranging from to in time domain.;t is the number of neighbouring frames used u0434u043bu00a0 temporal filtering, w (p, s, r) - weight u043fu0438u043au0441u0435u043bu00a0 inp2 (xn + p, yn + s, tn + (r), which u0432u044bu0447u0438u0441u043bu00a0u044eu0442 as;t is the adaptation of the filter to the power circuit, the u043du043eu0440u043cu0438u0440u043eu0432u043eu0447u043du044bu0439 multiplier u0432u044bu0447u0438u0441u043bu00a0u044eu0442 formula;.;4. method for u043eu0442u043bu0438u0447u0430u044eu0449u0438u0439u0441u00a0 1, so that the u0432u044bu0447u0438u0441u043bu00a0u044eu0442 result bilateral filtering through the weighted summation of the first and second u043fu043eu0434u043eu0431u043bu0430u0441u0442 filtering and, like,;where bil _ pixel denoting the result of adaptive u0431u0438u043bu0430u0442u0435u0440u0430u043bu044cu043du043eu0433u043e spatial filter;filt _ pixel1 denotes the result of first subarea.;filt _ pixel2 denotes the result of second subarea.;w1 weight result filtering first subarea u0432u044bu0447u0438u0441u043bu00a0u044eu0442 as follows;w2 weight result filtering second subarea u0432u044bu0447u0438u0441u043bu00a0u044eu0442 similarly as w1;where the sum is the sum of _ grad1 gradient, calculated along the contour of the first subarea, sum _ grad2 is the gradient, calculated along the path by the second u043fu043eu0434u043eu0431u043bu0430u0441 ti.;5. method for u043eu0442u043bu0438u0447u0430u044eu0449u0438u0439u0441u00a0 1, so that the final result as u0432u044bu0447u0438u0441u043bu00a0u044eu0442;filt _ pixel = wbil _ pixel + (1 - (w) int _ pixel.;where th is the threshold that u043eu043fu0440u0435u0434u0435u043bu00a0u0435u0442u0441u00a0 by u043eu0431u0443u0447u0435u043du0438u00a0 algorithm, bil _ pixel denoting the result of bilateral filtering, int _ pixel denoting the result of ams u0440u043eu043au0441u0438u043cu0430u0446u0438u0438 current u043fu0438u043au0441u0435u043bu00a0 along the contour, filt _ pixel refers to the final result.