healer-8

一.算法原理
1.定義：自動白平衡的目標是消除光源色溫對圖像顏色的影響，使白色物體在不同光照下都能呈現真實的白色。
2.改進點：
1）灰度世界全局處理導致局部色偏。
2）局部白平衡在顏色均勻區域產生的分塊效應。
3.算法設計：
1）局部處理：使用滑動窗口實現局部自適應，避免全局處理導致的色偏。
2）雙策略融合：結合“灰度世界”和“白點檢測”假設
3）重疊窗口：步長小于窗口大小，平滑過渡不同區域的增益。
二.算法結構與代碼

function outImg = AWB_Slide(inputImg, filterSize, weightGray)

    stepSize = 40;  %偶數同時可被4除盡，避免窗口落在錯誤位置

    img = double(inputImg);         
    [height, width] = size(img);    
    outImg = zeros(size(inputImg));  
    
    %增益矩陣
    gainR = zeros(height, width);   
    gainB = zeros(height, width);   
    weight = zeros(height, width);  

    for i = 1:stepSize:height
        for j = 1:stepSize:width

            %當前窗口
            rowStart = max(1, min(i, height-filterSize+1));
            colStart = max(1, min(j, width-filterSize+1));
            rowEnd = rowStart + filterSize - 1;
            colEnd = colStart + filterSize - 1;
            window = img(rowStart:rowEnd, colStart:colEnd);
            
            % 分離RGGB
            R = window(1:2:end, 1:2:end);   
            Gr = window(1:2:end, 2:2:end);  
            Gb = window(2:2:end, 1:2:end);  
            B = window(2:2:end, 2:2:end);   
            
            meanG = (mean(Gr(:)) + mean(Gb(:)))/2;
            maxG = max(max(Gr(:)), max(Gb(:)));
            
            % R增益
            meanR = mean(R(:));
            maxR = max(R(:));
            kR = meanG / meanR;           
            pR = maxG / maxR;         
            gainRVal = weightGray*kR + (1-weightGray)*pR;
            
            % B增益
            meanB = mean(B(:));
            maxB = max(B(:));
            kB = meanG / meanB;
            pB = maxG / maxB;
            gainBVal = weightGray*kB + (1-weightGray)*pB;

            %增益累計
            %R
            gainR(rowStart:2:rowEnd, colStart:2:colEnd) = ...
                gainR(rowStart:2:rowEnd, colStart:2:colEnd) + gainRVal;
            %B
            gainB(rowStart+1:2:rowEnd, colStart+1:2:colEnd) = ...
                gainB(rowStart+1:2:rowEnd, colStart+1:2:colEnd) + gainBVal;
            
            weight(rowStart:rowEnd, colStart:colEnd) = ...
                weight(rowStart:rowEnd, colStart:colEnd) + 1;
        end
    end
    
    % R
    outImg(1:2:height, 1:2:width) = img(1:2:height, 1:2:width) .* ...
        (gainR(1:2:height, 1:2:width) ./ weight(1:2:height, 1:2:width));
    % B
    outImg(2:2:height, 2:2:width) = img(2:2:height, 2:2:width) .* ...
        (gainB(2:2:height, 2:2:width) ./ weight(2:2:height, 2:2:width));

    outImg(1:2:height, 2:2:width) = img(1:2:height, 2:2:width);
    outImg(2:2:height, 1:2:width) = img(2:2:height, 1:2:width);
    outImg = uint16(outImg);
end

內存局限：
1.1幀原始Bayer數據（用于當前窗口處理）。
2.gainR、gainB、weight 矩陣各需1幀內存（存儲累加值）。