MSRCR

实现了车辆目标的背景建模、前景检测、阴影消除、车辆跟踪(Achieve a vehicle target background modeling, foreground detection, shadow elimination, vehicle tracking)

SURF算法作为一种新近出现的特征提取方法，在重复度、独特性、鲁棒性3个方面，均超越或接近以往提出的同类方法，并在计算效率上具有明显的优势。本代码采用SURF算法检测图像并进行坐标变换与图像拼接。 采用SURF算法对图像进行检测，其主要是用Hessian矩阵对图像进行检测，对图像的特征提取之后找到图像的特征点。之后采用最近临快速匹配（NN）、随机抽样一致性（RANSAC）算法和最小二乘法参数优化（LM）对特征点进行提纯匹配。最后在两幅图像中进行坐标变换，达到统一坐标系和图像拼接的效果。 (SURF（Speeded Up Robust Features） as a method of feature extraction which newly appeared is over or nearly previous method on duplication, uniqueness, and robustness and have a clear advantage on computational efficiency. This code uses the SURF and coordinate transformation algorithm to detect image and image matching. This code uses the SURF algorithm of image detection, the main is to use the Hessian matrix of image for testing, to find the image after image feature extraction of feature points. After we used Nearest Neighbor (NN), Random Sample Consensus (RANSAC) algorithm and least square parameter optimization for purification of matching feature points. Coordinate transformation in the last two images, to coordinate system to achieve the same image stitching. )

将matlab处理得到的三维矩阵转化为envi软件可读取的数据格式raw(change the 3-D data from matlab to the raw form ENVI can read)

运用超像素分割算法对图像进行预分割，再利用dbscan聚类算法进行聚类最终实现图像分割。(The use of ultra-pixel segmentation algorithm for image pre-segmentation, reuse dbscan clustering algorithm to cluster the final image segmentation.)

将原始载体图片经过一维冗余离散小波变换分成四个子频带，分别对每个子频带进行奇异值分解， 将水印图片灰度值直接按比例缩小嵌入到奇异值矩阵中，再将此矩阵进行一次奇异值分解得到奇异值当做此子带的奇异值， 重建此子带矩阵，而后将进行反冗余离散小波变换得到嵌入载体后的水印图像。(The original image after the one-dimensional vector redundant discrete wavelet transform is divided into four subbands, respectively, of each sub-band for the singular value decomposition, the watermark image gray value scale down directly embedded into the matrix singular value, then this matrix is ​ ​ a Singular value decomposition singular value as the singular values ​ ​ of this sub-band, the reconstruction of this sub-band matrix and after anti-redundant discrete wavelet transform watermark image after embedding the carrier.)

二维最大熵阈值分割法，较一维有很好的去噪效果(Two-dimensional maximum entropy threshold segmentation method, the more one-dimensional results have a good de-noising)

对图像分块,计算每个块的均值和方差,并给出了结果的图象化表示,易于观察(Image sub-blocks, each block calculated the mean and variance, and gives the results of the image that is easy to observe)

曲面拟合法是一种高精度、计算量小、抗噪性能强的方法，当前已被广泛地应用于 各种场景。比如岩石材料的变形测量、电子器件变形的测量、基于掌纹图像相关匹 配的重复定位技术等。曲面拟合法能够成功计算出亚像素位移的前提是整像素搜索阶 段可以正确搜索出模板的整像素匹配点，一旦整像素匹配点搜索出现错误，那么亚像素 测量阶段所得到的位移距离就会失去意义。 使用不同的拟合函数对相关系数矩阵进行曲面拟合，会对结果产生一定的影响，比 较常用的拟合函数有二次函数、三次函数、高斯函数等。(Surface fitting method is a high-precision, computational complexity is small, strong anti-noise performance method, has been widely used in various scenes. For example, deformation measurement of rock materials, deformation measurement of electronic devices, and repetitive location technology based on palmprint image correlation matching. Surface fitting method can successfully calculate the sub-pixel displacement on the premise that the whole pixel search stage can correctly search the template of the whole pixel matching point, once the whole pixel matching point search error, then the sub-pixel measurement phase of the displacement distance will be meaningless. Using different fitting functions to fit the surface of the correlation coefficient matrix will have a certain impact on the results. The commonly used fitting functions are quadratic function, cubic function, Gaussian function and so on.)

说明：   核密度估计，绘制频率直方图、核密度估计图、正态分布密度图(Kernel density estimation, drawing frequency histograms, kernel density estimation map, normal density map)

MATLAB可见水印、不可见水印添加程序源代码，可见水印部分采用中值法，不可见部分采用DWT变换。(MATLAB visible watermark, invisible watermark to add the source code, we can see the watermark part in value method, the invisible part on DWT.)