Digital-Image-Processing-writed-by-Gonzalez-secon

具体实现思路：以4位数字的验证码为例 1、人工将验证码的4位数字每位对应的代码存入数据库中，每位存入0-9对应的代码，每个数可以多存这样可以提高识别率; 2、获取验证码以后，对其进行去背景、灰度处理、去噪点处理、分片处理以后生成每位数字对应的代码; 3、去背景色，这一步的目的是把验证码和背景颜色区别开来。4、去噪声：这一步要取出图像上的孤立点。这些孤立点被认为是噪声。  孤立点的定义：某个点，周围没有与该点等值的点。  或者某个连接块，该连接块的元素的个数小于某个给定值K， 把元素个数很小的连接块也定义为孤立点，有助于去处噪声。  去噪声算法：参照去背景算法。 5、图像锐化：图像锐化的目的是增强边界。这一步是可选的。看验证码的情况，这一步可以跳过。 6、图片有效区域截取：这个操作是将图片除验证码字符以外的边框去掉，只留下验证码字符图片，这样保证分片的准确性。 7、图片分片处理，这个处理是将整个图片分割成单个字符图片。8、拿分片图像生成的代码(eg:000111110011…..)与DB中已有代码相比较，取得相似度最高的即要验证码的字符;   注：如果DB的样本越多，识别的准确率也越高，但是速度会相应变慢

说明：  使用Splitting算法和GLA算法实现对图像的矢量量化及重新编码和重建。(Vector Quantization:To generate a 16-codeword codebook using Splitting Algorithm and GLA,then use this codebook to code the training image and reconstuct it.)

用于矩形检测的源代码，已配说明，易看懂。但不适合于车牌检测等复杂的矩形检测(Rectangular detection for the source code has been assigned that easy to understand. But not suitable for detection of license plate detection rectangular complex)

wu-manber算法纯c实现，多模匹配算法，效率比较高，喜欢的可以下载看看，已经调试，没有问题(wu-manber algorithm)

利用稀疏表示进行图像去噪时，需要生成两种学习字典，该源码比较了生成的两种字典的优劣(Denoising sparse representation, it is necessary to generate two learning dictionary, which compares the generated source code of the merits of two dictionary)

说明：  随机森林的行人检测。针对视频和图像中快速、准确的行人检测问题,提出了一种分层次的、全局信息和局部信息相结合的行人检测算法。该方法以随机森林分类器为基础。(Pedestrian detection based on random forest)

先检测出椭圆，然后对椭圆进行椭圆拟合，画椭圆。(First detected in an oval, and then carried out on the elliptical ellipse fitting, painted oval.)

说明：  通过hough变换求图像中的圆的参数（圆心坐标和半径）(Through the hough transform for image circle parameters (center coordinates and radius))

背景减除的完整算法，《A Statistical Approach for Real-time Robust Background Subtraction and Shadow Detection》作者源代码(The "src" folder contains all of the source code. To run the program, follow these simple steps: 1. Start MATLAB and ensure the "src" folder is your current working directory. 2. Edit "bs.m" and "classify.m" so that the file paths point to the proper directories as specified. 3. In the MATLAB environment type "bs" which will train the background model from the sequence you specified in (2). 4. In the MATLAB environment type "classify" which will classify the sequence of images you specified in (2). 5. Now the classified images should be saved to disk in the folder you specified in (2).)

基于小波变换的使用SPIHT算法进行图像压缩，并对原有的算法进行优化(Based on wavelet transform using the SPIHT algorithm for image compression, and the original algorithm to optimize)