/// <summary>
/// 实现功能： 基于Frequency-tuned 的图像显著性检测
/// 参考论文： Frequency-tuned Salient Region Detection， Radhakrishna Achantay， Page 4-5, 2009 CVPR
/// http://ivrgwww.epfl.ch/supplementary_material/RK_CVPR09/
/// 参考论文作者RGB转LAB的浮点版本
/// </summary>
/// <param name="srcBitmap"></param>
/// <returns></returns>
public static Bitmap SalientRegionDetectionBasedOnFT(Bitmap srcBitmap)
{
int width = srcBitmap.Width, height = srcBitmap.Height;
Rectangle rect = new Rectangle(0, 0, width, height);
BitmapData srcBmData = srcBitmap.LockBits(rect,
ImageLockMode.ReadWrite, PixelFormat.Format24bppRgb);
Bitmap dstBitmap = BasicMethodClass.CreateGrayscaleImage(width, height);
BitmapData dstBmData = dstBitmap.LockBits(rect,
ImageLockMode.ReadWrite, PixelFormat.Format8bppIndexed);
IntPtr srcScan = srcBmData.Scan0;
IntPtr dstScan = dstBmData.Scan0;
unsafe
{
double * srcP = (double *)(void *)srcScan;
byte * dstP = (byte *)(void *)dstScan;
int Index = 0, CurIndex = 0, srcStride = srcBmData.Stride, dstStride = dstBmData.Stride, X, Y;
double MeanL = 0, MeanA = 0, MeanB = 0;
double[] LabF;
double[] DistMap = new double[dstStride * height];
LabF = BasicMethodClass.sRGBtoXYZ(srcBmData, width, height, out MeanL, out MeanA, out MeanB);
// LabF = BasicMethodClass.GaussianSmooth(LabF, width, srcBmData.Stride, srcBmData.Height);
double maxval = 0;
double minval = double.MaxValue;
int i = 0;
for (Y = 0; Y < height; Y++)
{
Index = Y * srcStride;
CurIndex = Y * dstStride; //
for (X = 0; X < width; X++) // 计算像素的显著性
{ //此处有时会出现数组越界，原因暂时未知(已解决，目标8位图有3字节的对齐)
double curValue = (MeanL - LabF[Index]) * (MeanL - LabF[Index]) + (MeanA - LabF[Index + 1]) * (MeanA - LabF[Index + 1]) + (MeanB - LabF[Index + 2]) * (MeanB - LabF[Index + 2]);
DistMap[CurIndex] = curValue;
Index += 3;
CurIndex++;
if (maxval < curValue)
{
maxval = curValue;
}
if (minval > curValue)
{
minval = curValue;
}
i++;
}
}
BasicMethodClass.Normalize(DistMap, dstBmData.Stride, height, maxval, minval, dstP);
//写入序列化函数中。。
}
srcBitmap.UnlockBits(srcBmData);
dstBitmap.UnlockBits(dstBmData);
return(dstBitmap);
}
public static Bitmap SalientRegionDetectionBasedOnFT(Bitmap srcBitmap, float distcof)
{
int width = srcBitmap.Width, height = srcBitmap.Height;
Rectangle rect = new Rectangle(0, 0, width, height);
BitmapData srcBmData = srcBitmap.LockBits(rect,
ImageLockMode.ReadWrite, PixelFormat.Format24bppRgb);
Bitmap dstBitmap = BasicMethodClass.CreateGrayscaleImage(width, height);
BitmapData dstBmData = dstBitmap.LockBits(rect,
ImageLockMode.ReadWrite, PixelFormat.Format8bppIndexed);
IntPtr srcScan = srcBmData.Scan0;
IntPtr dstScan = dstBmData.Scan0;
unsafe
{
double * srcP = (double *)(void *)srcScan;
byte * dstP = (byte *)(void *)dstScan;
int Index = 0, CurIndex = 0, srcStride = srcBmData.Stride, dstStride = dstBmData.Stride, X, Y;
double MeanL = 0, MeanA = 0, MeanB = 0;
double[] LabF;
double[] DistMap = new double[dstStride * height];
LabF = BasicMethodClass.sRGBtoXYZ(srcBmData, width, height, out MeanL, out MeanA, out MeanB);
// LabF = BasicMethodClass.GaussianSmooth(LabF, width, srcBmData.Stride, srcBmData.Height);
//加入显著位置权重
int wk = (int)(width / 2), hk = (int)(height / 2);
for (Y = 0; Y < height; Y++)
{
Index = Y * srcStride;
CurIndex = Y * dstStride; //
for (X = 0; X < width; X++) // 计算像素的显著性
{ //此处有时会出现数组越界，原因暂时未知(已解决，目标8位图有3字节的对齐)
DistMap[CurIndex] = (MeanL - LabF[Index]) * (MeanL - LabF[Index]) + (MeanA - LabF[Index + 1]) * (MeanA - LabF[Index + 1]) + (MeanB - LabF[Index + 2]) * (MeanB - LabF[Index + 2]);
DistMap[CurIndex] *= Math.Max(1 - (Math.Abs(wk - X) + Math.Abs(hk - Y)) / (wk + hk + 1.0), distcof);
Index += 3;
CurIndex++;
}
}
DistMap = BasicMethodClass.Normalize(DistMap, dstBmData.Stride, height, dstP);
//写入序列化函数中。。
}
srcBitmap.UnlockBits(srcBmData);
dstBitmap.UnlockBits(dstBmData);
return(dstBitmap);
}
/// <summary>
/// 实现功能： saliency is determined as the local contrast of an image region with respect to its neighborhood at various scales
/// 参考论文： Salient Region Detection and Segmentation Radhakrishna Achanta, Francisco Estrada, Patricia Wils, and Sabine SÄusstrunk 2008 , Page 4-5
/// 参考laviewpbt的C实现 http://blog.csdn.net/laviewpbt/article/details/38357017
/// </summary>
/// <param name="srcBitmap"></param>
/// <returns></returns>
public static Bitmap SalientRegionDetectionBasedonLC(Bitmap srcBitmap)
{
int width = srcBitmap.Width, height = srcBitmap.Height;
Rectangle rect = new Rectangle(0, 0, width, height);
BitmapData srcBmData = srcBitmap.LockBits(rect,
ImageLockMode.ReadWrite, PixelFormat.Format24bppRgb);
Bitmap dstBitmap = BasicMethodClass.CreateGrayscaleImage(width, height);
BitmapData dstBmData = dstBitmap.LockBits(rect,
ImageLockMode.ReadWrite, PixelFormat.Format8bppIndexed);
IntPtr srcScan = srcBmData.Scan0;
IntPtr dstScan = dstBmData.Scan0;
unsafe
{
byte *srcP = (byte *)(void *)srcScan;
byte *dstP = (byte *)(void *)dstScan;
int Value;
int DistMaxValue = 0; //用于记录最大距离，用于最后归纳图片的有效颜色范围
int IndexBit, CurIndex; //用于定位字节位和像素位置
int[] Gray = new int[width * height]; //用于保存计算完成了的灰度值
int[] HistGram = new int[256]; //用于直方图统计
int[] Dist = new int[256]; //用于记录每种颜色到各个颜色的距离
//int[] DistMap = new int[width*height];//显著性图
//创建直方图和灰度图
for (int y = 0; y < height; y++)
{
IndexBit = srcBmData.Stride * y;
CurIndex = width * y;
for (int x = 0; x < width; x++)
{
Value = (srcP[IndexBit] + srcP[IndexBit + 1] * 2 + srcP[IndexBit + 2]) / 4;
HistGram[Value]++;
Gray[CurIndex] = Value;
IndexBit += 3;
CurIndex++;
}
}
//记录颜色距离
for (int Y = 0; Y < 256; Y++)
{
Value = 0;
for (int X = 0; X < 256; X++)
{
Value += Math.Abs(Y - X) * HistGram[X];//Y点灰度与整张图的距离
}
Dist[Y] = Value;
if (DistMaxValue < Value)
{
DistMaxValue = Value;
}
}
//计算用于减小距离为有效颜色值的倍数
DistMaxValue = DistMaxValue / 256;
//计算显著性图片
for (int Y = 0; Y < height; Y++)
{
CurIndex = Y * width;
IndexBit = Y * dstBmData.Stride;
for (int X = 0; X < width; X++)
{
Value = Dist[Gray[CurIndex]]; // 计算全图每个像素的显著性
dstP[IndexBit] = (byte)(Value / DistMaxValue);
CurIndex++;
IndexBit++;
}
}
}
srcBitmap.UnlockBits(srcBmData);
dstBitmap.UnlockBits(dstBmData);
return(dstBitmap);
}
