昨天看了阮一峰老师的文章:《相似图片搜索原理》,于是把直方图和向量那块算法用js实现了一下,
源码如下:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 | function getHistogram(imageData) { var arr = []; for (var i = 0; i < 64; i++) { arr[i] = 0; } var data = imageData.data; var pow4 = Math.pow(4, 2); for (var i = 0, len = data.length; i < len; i += 4) { var red = (data[i] / 64) | 0; var green = (data[i + 1] / 64) | 0; var blue = (data[i + 2] / 64) | 0; var index = red * pow4 + green * 4 + blue; arr[index]++; } return arr; } function cosine(arr1, arr2) { var axb = 0, a = 0, b = 0; for (var i = 0, len = arr1.length; i < len; i++) { axb += arr1[i] * arr2[i]; a += arr1[i] * arr1[i]; b += arr2[i] * arr2[i]; } return axb / (Math.sqrt(a) * Math.sqrt(b)); } function gray(imgData) { var data = imgData.data; for (var i = 0, len = data.length; i < len; i += 4) { var gray = parseInt((data[i] + data[i + 1] + data[i + 2]) / 3); data[i + 2] = data[i + 1] = data[i] = gray; } return imgData; } |
有个问题,假如图片是灰色的跟原图进行比较,那么要比较相似度,需要将图片都转成灰色的,即使用上面代码的gray函数来处理