java 图片相似度算法
转载自：https://www.sunjs.com/article/detail/24dd9a9e436e489185430c4c45034d69.html
利用直方图原理实现图像内容相似度比较、均值哈希实现图像内容相似度比较、汉明距离算法实现图像内容相似度比较
直方图原理实现图像内容相似度比较算法：
import javax.imageio.*;
import java.awt.image.*;
import java.awt.*;
import java.io.*;
public class PhotoDigest {
public static void main(String[] args) throws Exception {
float percent = compare(getData("/Users/sun/Downloads/1.jpg"),
getData("/Users/sun/Downloads/2.jpg"));
if (percent == 0) {
System.out.println("无法比较");
} else {
System.out.println("两张图片的相似度为：" + percent + "%");
}
}
public static int[] getData(String name) {
try {
BufferedImage img = ImageIO.read(new File(name));
BufferedImage slt = new BufferedImage(100, 100,
BufferedImage.TYPE_INT_RGB);
slt.getGraphics().drawImage(img, 0, 0, 100, 100, null);
// ImageIO.write(slt,"jpeg",new File("slt.jpg"));
int[] data = new int[256];
for (int x = 0; x < slt.getWidth(); x++) {
for (int y = 0; y < slt.getHeight(); y++) {
int rgb = slt.getRGB(x, y);
Color myColor = new Color(rgb);
int r = myColor.getRed();
int g = myColor.getGreen();
int b = myColor.getBlue();
data[(r + g + b) / 3]++;
}
}
// data 就是所谓图形学当中的直方图的概念
return data;
} catch (Exception exception) {
System.out.println("有文件没有找到,请检查文件是否存在或路径是否正确");
return null;
}
}
public static float compare(int[] s, int[] t) {
try {
float result = 0F;
for (int i = 0; i < 256; i++) {
int abs = Math.abs(s[i] - t[i]);
int max = Math.max(s[i], t[i]);
result += (1 - ((float) abs / (max == 0 ? 1 : max)));
}
return (result / 256) * 100;
} catch (Exception exception) {
return 0;
}
}
}
均值哈希实现图像内容相似度比较算法：
import java.awt.Graphics;
import java.awt.Image;
import java.awt.color.ColorSpace;
import java.awt.image.BufferedImage;
import java.awt.image.ColorConvertOp;
import java.io.File;
import java.io.IOException;
import java.util.Arrays;
import javax.imageio.ImageIO;
/**
* 均值哈希实现图像指纹比较
*
*/
public final class FingerPrint {
public static void main(String[] args) {
try {
FingerPrint fp1 = new FingerPrint(ImageIO.read(new File(
"/Users/sun/Downloads/1.jpg")));
FingerPrint fp2 = new FingerPrint(ImageIO.read(new File(
"/Users/sun/Downloads/2.jpg")));
System.out.println(fp1.toString(true));
System.out.printf("sim=%f", fp1.compare(fp2));
} catch (IOException e) {
e.printStackTrace();
}
}
/**
* 图像指纹的尺寸,将图像resize到指定的尺寸，来计算哈希数组
*/
private static final int HASH_SIZE = 16;
/**
* 保存图像指纹的二值化矩阵
*/
private final byte[] binaryzationMatrix;
public FingerPrint(byte[] hashValue) {
if (hashValue.length != HASH_SIZE * HASH_SIZE)
throw new IllegalArgumentException(String.format(
"length of hashValue must be %d", HASH_SIZE * HASH_SIZE));
this.binaryzationMatrix = hashValue;
}
public FingerPrint(String hashValue) {
this(toBytes(hashValue));
}
public FingerPrint(BufferedImage src) {
this(hashValue(src));
}
private static byte[] hashValue(BufferedImage src) {
BufferedImage hashImage = resize(src, HASH_SIZE, HASH_SIZE);
byte[] matrixGray = (byte[]) toGray(hashImage).getData()
.getDataElements(0, 0, HASH_SIZE, HASH_SIZE, null);
return binaryzation(matrixGray);
}
/**
* 从压缩格式指纹创建{@link FingerPrint}对象
*
* @param compactValue
* @return
*/
public static FingerPrint createFromCompact(byte[] compactValue) {
return new FingerPrint(uncompact(compactValue));
}
public static boolean validHashValue(byte[] hashValue) {
if (hashValue.length != HASH_SIZE)
return false;
for (byte b : hashValue) {
if (0 != b && 1 != b)
return false;
}
return true;
}
public static boolean validHashValue(String hashValue) {
if (hashValue.length() != HASH_SIZE)
return false;
for (int i = 0; i < hashValue.length(); ++i) {
if ('0' != hashValue.charAt(i) && '1' != hashValue.charAt(i))
return false;
}
return true;
}
public byte[] compact() {
return compact(binaryzationMatrix);
}
/**
* 指纹数据按位压缩
*
* @param hashValue
* @return
*/
private static byte[] compact(byte[] hashValue) {
byte[] result = new byte[(hashValue.length + 7) >> 3];
byte b = 0;
for (int i = 0; i < hashValue.length; ++i) {
if (0 == (i & 7)) {
b = 0;
}
if (1 == hashValue[i]) {
b |= 1 << (i & 7);
} else if (hashValue[i] != 0)
throw new IllegalArgumentException(
"invalid hashValue,every element must be 0 or 1");
if (7 == (i & 7) || i == hashValue.length - 1) {
result[i >> 3] = b;
}
}
return result;
}
/**
* 压缩格式的指纹解压缩
*
* @param compactValue
* @return
*/
private static byte[] uncompact(byte[] compactValue) {
byte[] result = new byte[compactValue.length << 3];
for (int i = 0; i < result.length; ++i) {
if ((compactValue[i >> 3] & (1 << (i & 7))) == 0)
result[i] = 0;
else
result[i] = 1;
}
return result;
}
/**
* 字符串类型的指纹数据转为字节数组
*
* @param hashValue
* @return
*/
private static byte[] toBytes(String hashValue) {
hashValue = hashValue.replaceAll("\\s", "");
byte[] result = new byte[hashValue.length()];
for (int i = 0; i < result.length; ++i) {
char c = hashValue.charAt(i);
if ('0' == c)
result[i] = 0;
else if ('1' == c)
result[i] = 1;
else
throw new IllegalArgumentException("invalid hashValue String");
}
return result;
}
/**
* 缩放图像到指定尺寸
*
* @param src
* @param width
* @param height
* @return
*/
private static BufferedImage resize(Image src, int width, int height) {
BufferedImage result = new BufferedImage(width, height,
BufferedImage.TYPE_3BYTE_BGR);
Graphics g = result.getGraphics();
try {
g.drawImage(
src.getScaledInstance(width, height, Image.SCALE_SMOOTH),
0, 0, null);
} finally {
g.dispose();
}
return result;
}
/**
* 计算均值
*
* @param src
* @return
*/
private static int mean(byte[] src) {
long sum = 0;
// 将数组元素转为无符号整数
for (byte b : src)
sum += (long) b & 0xff;
return (int) (Math.round((float) sum / src.length));
}
/**
* 二值化处理
*
* @param src
* @return
*/
private static byte[] binaryzation(byte[] src) {
byte[] dst = src.clone();
int mean = mean(src);
for (int i = 0; i < dst.length; ++i) {
// 将数组元素转为无符号整数再比较
dst[i] = (byte) (((int) dst[i] & 0xff) >= mean ? 1 : 0);
}
return dst;
}
/**
* 转灰度图像
*
* @param src
* @return
*/
private static BufferedImage toGray(BufferedImage src) {
if (src.getType() == BufferedImage.TYPE_BYTE_GRAY) {
return src;
} else {
// 图像转灰
BufferedImage grayImage = new BufferedImage(src.getWidth(),
src.getHeight(), BufferedImage.TYPE_BYTE_GRAY);
new ColorConvertOp(ColorSpace.getInstance(ColorSpace.CS_GRAY), null)
.filter(src, grayImage);
return grayImage;
}
}
@Override
public String toString() {
return toString(true);
}
/**
* @param multiLine
* 是否分行
* @return
*/
public String toString(boolean multiLine) {
StringBuffer buffer = new StringBuffer();
int count = 0;
for (byte b : this.binaryzationMatrix) {
buffer.append(0 == b ? '0' : '1');
if (multiLine && ++count % HASH_SIZE == 0)
buffer.append('\n');
}
return buffer.toString();
}
@Override
public boolean equals(Object obj) {
if (obj instanceof FingerPrint) {
return Arrays.equals(this.binaryzationMatrix,
((FingerPrint) obj).binaryzationMatrix);
} else
return super.equals(obj);
}
/**
* 与指定的压缩格式指纹比较相似度
*
* @param compactValue
* @return
* @see #compare(FingerPrint)
*/
public float compareCompact(byte[] compactValue) {
return compare(createFromCompact(compactValue));
}
/**
* @param hashValue
* @return
* @see #compare(FingerPrint)
*/
public float compare(String hashValue) {
return compare(new FingerPrint(hashValue));
}
/**
* 与指定的指纹比较相似度
*
* @param hashValue
* @return
* @see #compare(FingerPrint)
*/
public float compare(byte[] hashValue) {
return compare(new FingerPrint(hashValue));
}
/**
* 与指定图像比较相似度
*
* @param image2
* @return
* @see #compare(FingerPrint)
*/
public float compare(BufferedImage image2) {
return compare(new FingerPrint(image2));
}
/**
* 比较指纹相似度
*
* @param src
* @return
* @see #compare(byte[], byte[])
*/
public float compare(FingerPrint src) {
if (src.binaryzationMatrix.length != this.binaryzationMatrix.length)
throw new IllegalArgumentException(
"length of hashValue is mismatch");
return compare(binaryzationMatrix, src.binaryzationMatrix);
}
/**
* 判断两个数组相似度，数组长度必须一致否则抛出异常
*
* @param f1
* @param f2
* @return 返回相似度(0.0~1.0)
*/
private static float compare(byte[] f1, byte[] f2) {
if (f1.length != f2.length)
throw new IllegalArgumentException("mismatch FingerPrint length");
int sameCount = 0;
for (int i = 0; i < f1.length; ++i) {
if (f1[i] == f2[i])
++sameCount;
}
return (float) sameCount / f1.length;
}
public static float compareCompact(byte[] f1, byte[] f2) {
return compare(uncompact(f1), uncompact(f2));
}
public static float compare(BufferedImage image1, BufferedImage image2) {
return new FingerPrint(image1).compare(new FingerPrint(image2));
}
}
汉明距离算法实现图像内容相似度比较算法：
import java.awt.Graphics2D;
import java.awt.color.ColorSpace;
import java.awt.image.BufferedImage;
import java.awt.image.ColorConvertOp;
import java.io.File;
import java.io.FileInputStream;
import java.io.FileNotFoundException;
import java.io.InputStream;
import javax.imageio.ImageIO;
/*
* pHash-like image hash.
* Author: Elliot Shepherd ([email protected]
* Based On: http://www.hackerfactor.com/blog/index.php?/archives/432-Looks-Like-It.html
*/
public class ImagePHash {
private int size = 32;
private int smallerSize = 8;
public ImagePHash() {
initCoefficients();
}
public ImagePHash(int size, int smallerSize) {
this.size = size;
this.smallerSize = smallerSize;
initCoefficients();
}
public int distance(String s1, String s2) {
int counter = 0;
for (int k = 0; k < s1.length(); k++) {
if (s1.charAt(k) != s2.charAt(k)) {
counter++;
}
}
return counter;
}
// Returns a 'binary string' (like. 001010111011100010) which is easy to do
// a hamming distance on.
public String getHash(InputStream is) throws Exception {
BufferedImage img = ImageIO.read(is);
/*
* 1. Reduce size. Like Average Hash, pHash starts with a small image.
* However, the image is larger than 8x8; 32x32 is a good size. This is
* really done to simplify the DCT computation and not because it is
* needed to reduce the high frequencies.
*/
img = resize(img, size, size);
/*
* 2. Reduce color. The image is reduced to a grayscale just to further
* simplify the number of computations.
*/
img = grayscale(img);
double[][] vals = new double[size][size];
for (int x = 0; x < img.getWidth(); x++) {
for (int y = 0; y < img.getHeight(); y++) {
vals[x][y] = getBlue(img, x, y);
}
}
/*
* 3. Compute the DCT. The DCT separates the image into a collection of
* frequencies and scalars. While JPEG uses an 8x8 DCT, this algorithm
* uses a 32x32 DCT.
*/
long start = System.currentTimeMillis();
double[][] dctVals = applyDCT(vals);
System.out.println("DCT: " + (System.currentTimeMillis() - start));
/*
* 4. Reduce the DCT. This is the magic step. While the DCT is 32x32,
* just keep the top-left 8x8. Those represent the lowest frequencies in
* the picture.
*/
/*
* 5. Compute the average value. Like the Average Hash, compute the mean
* DCT value (using only the 8x8 DCT low-frequency values and excluding
* the first term since the DC coefficient can be significantly
* different from the other values and will throw off the average).
*/
double total = 0;
for (int x = 0; x < smallerSize; x++) {
for (int y = 0; y < smallerSize; y++) {
total += dctVals[x][y];
}
}
total -= dctVals[0][0];
double avg = total / (double) ((smallerSize * smallerSize) - 1);
/*
* 6. Further reduce the DCT. This is the magic step. Set the 64 hash
* bits to 0 or 1 depending on whether each of the 64 DCT values is
* above or below the average value. The result doesn't tell us the
* actual low frequencies; it just tells us the very-rough relative
* scale of the frequencies to the mean. The result will not vary as
* long as the overall structure of the image remains the same; this can
* survive gamma and color histogram adjustments without a problem.
*/
String hash = "";
for (int x = 0; x < smallerSize; x++) {
for (int y = 0; y < smallerSize; y++) {
if (x != 0 && y != 0) {
hash += (dctVals[x][y] > avg ? "1" : "0");
}
}
}
return hash;
}
private BufferedImage resize(BufferedImage image, int width, int height) {
BufferedImage resizedImage = new BufferedImage(width, height,
BufferedImage.TYPE_INT_ARGB);
Graphics2D g = resizedImage.createGraphics();
g.drawImage(image, 0, 0, width, height, null);
g.dispose();
return resizedImage;
}
private ColorConvertOp colorConvert = new ColorConvertOp(
ColorSpace.getInstance(ColorSpace.CS_GRAY), null);
private BufferedImage grayscale(BufferedImage img) {
colorConvert.filter(img, img);
return img;
}
private static int getBlue(BufferedImage img, int x, int y) {
return (img.getRGB(x, y)) & 0xff;
}
// DCT function stolen from
// http://stackoverflow.com/questions/4240490/problems-with-dct-and-idct-algorithm-in-java
private double[] c;
private void initCoefficients() {
c = new double[size];
for (int i = 1; i < size; i++) {
c[i] = 1;
}
c[0] = 1 / Math.sqrt(2.0);
}
private double[][] applyDCT(double[][] f) {
int N = size;
double[][] F = new double[N][N];
for (int u = 0; u < N; u++) {
for (int v = 0; v < N; v++) {
double sum = 0.0;
for (int i = 0; i < N; i++) {
for (int j = 0; j < N; j++) {
sum += Math
.cos(((2 * i + 1) / (2.0 * N)) * u * Math.PI)
* Math.cos(((2 * j + 1) / (2.0 * N)) * v
* Math.PI) * (f[i][j]);
}
}
sum *= ((c[u] * c[v]) / 4.0);
F[u][v] = sum;
}
}
return F;
}
public static void main(String[] args) {
ImagePHash p = new ImagePHash();
String image1;
String image2;
try {
image1 = p.getHash(new FileInputStream(new File(
"/Users/sun/Downloads/1.jpg")));
image2 = p.getHash(new FileInputStream(new File(
"/Users/sun/Downloads/11.png")));
System.out.println("1:1 Score is " + p.distance(image1, image2));
} catch (FileNotFoundException e) {
e.printStackTrace();
} catch (Exception e) {
e.printStackTrace();
}
}
}
结果说明：汉明距离越大表明图片差异越大，如果不相同的数据位不超过5，就说明两张图片很相似；如果大于10，就说明这是两张不同的图片。
《三种基于感知哈希算法的相似图像检索技术》
根据Neal Krawetz博士的解释，原理非常简单易懂。我们可以用一个快速算法，就达到基本的效果。
这里的关键技术叫做"感知哈希算法"（Perceptual hash algorithm），它的作用是对每张图片生成一个"指纹"（fingerprint）字符串，然后比较不同图片的指纹。结果越接近，就说明图片越相似。
下面是一个最简单的实现：
第一步，缩小尺寸。
将图片缩小到8x8的尺寸，总共64个像素。这一步的作用是去除图片的细节，只保留结构、明暗等基本信息，摒弃不同尺寸、比例带来的图片差异。
第二步，简化色彩。
将缩小后的图片，转为64级灰度。也就是说，所有像素点总共只有64种颜色。
第三步，计算平均值。
计算所有64个像素的灰度平均值。
第四步，比较像素的灰度。
将每个像素的灰度，与平均值进行比较。大于或等于平均值，记为1；小于平均值，记为0。
第五步，计算哈希值。
将上一步的比较结果，组合在一起，就构成了一个64位的整数，这就是这张图片的指纹。组合的次序并不重要，只要保证所有图片都采用同样次序就行了。
得到指纹以后，就可以对比不同的图片，看看64位中有多少位是不一样的。在理论上，这等同于计算"汉明距离"（Hamming distance）。如果不相同的数据位不超过5，就说明两张图片很相似；如果大于10，就说明这是两张不同的图片。
具体的代码实现，可以参见Wote用python语言写的imgHash.py。代码很短，只有53行。使用的时候，第一个参数是基准图片，第二个参数是用来比较的其他图片所在的目录，返回结果是两张图片之间不相同的数据位数量（汉明距离）。
这种算法的优点是简单快速，不受图片大小缩放的影响，缺点是图片的内容不能变更。如果在图片上加几个文字，它就认不出来了。所以，它的最佳用途是根据缩略图，找出原图。
实际应用中，往往采用更强大的pHash算法和SIFT算法，它们能够识别图片的变形。只要变形程度不超过25%，它们就能匹配原图。这些算法虽然更复杂，但是原理与上面的简便算法是一样的，就是先将图片转化成Hash字符串，然后再进行比较。