图像处理之图像快速插值放缩算法

算法思想：

基于双线性算法的分解，分别进行水平与垂直两个方向的放缩，完成对整张图像的放大或

者缩小。基于的数学思想为矩阵的乘法，对一个scale矩阵可以拆分为水平与垂直方向的两

个关联矩阵，具体如下：

关于什么是双线性插值参加这里：http://blog.csdn.net/jia20003/article/details/6915185

程序解释：

类ScaleFilter完成对图像的快速放大与缩小，接受输入参数为XY方向的放缩比例值。

hscal, vscale的默认值为1.5f即将输入图像在XY放大1.5倍。XY方向的Scale方法参考与运用

了移动窗口的算法。感兴趣可以自己研究，我也是改写一段c语言代码得到。感觉非常的精妙。

程序效果如下：

Scale Filter的源代码如下：

package com.gloomyfish.filter.study;/** * @author gloomyfish * @date 2012-09-23 * @BLOGPAGE:http://blog.csdn.net/jia20003 */import java.awt.image.BufferedImage;import java.awt.image.ColorModel;public class ScaleFilter extends AbstractBufferedImageOp {/** * default will zoom in 2.0 * input size of original image. */private float hscale = 1.5f;private float vscale = 1.5f;public ScaleFilter() {}public void setHscale(float hscale) {this.hscale = hscale;}public void setVscale(float vscale) {this.vscale = vscale;}@Overridepublic BufferedImage filter(BufferedImage src, BufferedImage dest) {int width = src.getWidth();        int height = src.getHeight();        if ( dest == null )        dest = createCompatibleDestImage( src, null );        // initialization pixel data        int[] inPixels = new int[width*height];        int outwidth = (int)(hscale * (float)width);        int outheight = (int)(vscale * (float)height);        int[] outhPixels = new int[outwidth*height];        int[] outPixels = new int[outwidth*outheight];        // start to zoom in/out here        getRGB( src, 0, 0, width, height, inPixels );        hscale(inPixels, outhPixels, width, height);        vscale(outhPixels, outPixels, outwidth, height);        // create buffered image and return it with result image data         setRGB( dest, 0, 0, outwidth, outheight, outPixels);        return dest;}    public BufferedImage createCompatibleDestImage(BufferedImage src, ColorModel dstCM) {        if ( dstCM == null )            dstCM = src.getColorModel();        int outwidth = (int)(hscale * (float)src.getWidth());        int outheight = (int)(vscale * (float)src.getHeight());        return new BufferedImage(dstCM, dstCM.createCompatibleWritableRaster(outwidth, outheight), dstCM.isAlphaPremultiplied(), null);    }private void hscale(int[] input, int[] output, int width, int height) {int ta1 = 0, tr1 = 0, tg1 = 0, tb1 = 0;int ta2 = 0, tr2 = 0, tg2 = 0, tb2 = 0;int sumred = 0, sumgreen = 0, sumblue = 0;double accred = 0, accgreen = 0, accblue = 0;int p, q;int outwidth = (int)(this.hscale * width);double area = (outwidth * width);int inCol = 0, outCol = 0;int inIndex1 = 0, inIndex2 = 0, outIndex = 0;for (int row = 0; row < height; row++) {q = width;p = outwidth;accred = accgreen = accblue = 0;inCol = outCol = 0;while (outCol < outwidth) {if(outCol == 299) {System.out.println("what are you doing...");}if ((inCol + 1) < 2) {inIndex1 = row * width + inCol;inIndex2 = row * width + (inCol + 1);        ta1 = (input[inIndex1] >> 24) & 0xff;                tr1 = (input[inIndex1] >> 16) & 0xff;                tg1 = (input[inIndex1] >> 8) & 0xff;                tb1 = input[inIndex1] & 0xff;                        ta2 = (input[inIndex2] >> 24) & 0xff;                tr2 = (input[inIndex2] >> 16) & 0xff;                tg2 = (input[inIndex2] >> 8) & 0xff;                tb2 = input[inIndex2] & 0xff;                sumred = p * tr1 + (outwidth - p) * tr2;                sumgreen = p * tg1 + (outwidth - p) * tg2;                sumblue = p * tb1 + (outwidth - p) * tb2;}else {inIndex1 = row * width + inCol;        ta1 = (input[inIndex1] >> 24) & 0xff;                tr1 = (input[inIndex1] >> 16) & 0xff;                tg1 = (input[inIndex1] >> 8) & 0xff;                tb1 = input[inIndex1] & 0xff;                sumred = outwidth * tr1;                sumgreen = outwidth * tg1;                sumblue = outwidth * tb1;}if (p < q) {accred += sumred * p;accgreen += sumgreen * p;accblue += sumblue * p;q -= p;p = outwidth;inCol++;} else {accred += sumred * q;accgreen += sumgreen * q;accblue += sumblue * q;outIndex = row * outwidth + outCol;output[outIndex] = ta1 << 24 | ((int)(accred / area) << 16) | ((int)(accgreen / area) << 8) | (int)(accblue / area);accred = accgreen = accblue = 0;p -= q;q = width;outCol++;}}}}private void vscale(int[] input, int[] output, int width, int height) {int ta1 = 0, tr1 = 0, tg1 = 0, tb1 = 0;int ta2 = 0, tr2 = 0, tg2 = 0, tb2 = 0;int sumred = 0, sumgreen = 0, sumblue = 0;double accred = 0, accgreen = 0, accblue = 0;int inRow = 0, outRow = 0;int inIndex1 = 0, inIndex2 = 0, outIndex = 0;int p, q;int ih = height;int oh = (int)(height * vscale);int area = (ih * oh);for (int col = 0; col < width; col++) {q = ih;p = oh;accred = accgreen = accblue = 0;inRow = outRow = 0;while (outRow < oh) {if (inRow+1 < ih) {inIndex1 = inRow * width + col;inIndex2 = (inRow+1) * width + col;        ta1 = (input[inIndex1] >> 24) & 0xff;                tr1 = (input[inIndex1] >> 16) & 0xff;                tg1 = (input[inIndex1] >> 8) & 0xff;                tb1 = input[inIndex1] & 0xff;                        ta2 = (input[inIndex2] >> 24) & 0xff;                tr2 = (input[inIndex2] >> 16) & 0xff;                tg2 = (input[inIndex2] >> 8) & 0xff;                tb2 = input[inIndex2] & 0xff;                sumred = p * tr1 + (oh - p) * tr2;                sumgreen = p * tg1 + (oh - p) * tg2;                sumblue = p * tb1 + (oh - p) * tb2;}else{inIndex1 = inRow * width + col;        ta1 = (input[inIndex1] >> 24) & 0xff;                tr1 = (input[inIndex1] >> 16) & 0xff;                tg1 = (input[inIndex1] >> 8) & 0xff;                tb1 = input[inIndex1] & 0xff;                sumred = oh * tr1;                sumgreen = oh * tg1;                sumblue = oh * tb1;}if (p < q) {accred += sumred * p;accgreen += sumgreen * p;accblue += sumblue * p;q -= p;p = oh;inRow++;} else {accred += sumred * q;accgreen += sumgreen * q;accblue += sumblue * q;outIndex = outRow * width + col;output[outIndex] = ta1 << 24 | ((int)(accred / area) << 16) | ((int)(accgreen / area) << 8) | (int)(accblue / area);accred = accgreen = accblue = 0;p -= q;q = ih;outRow++;}}}}}

后注：其效果近似与双线性内插值算法，但是运行速度却是它的几十倍之多，感兴趣者

可以自己测试。证明我没有信口开河。