Description: 思路简要说明：
1、图像二值化
将图片中的各点用0或1表示，1为有效点，0为背景。这里使用的是最大类间方差法
（otsu），在资料中有介绍。
2、去除干扰点
3、分割
将整个的图片分为每个单独的字，在下一步中才能一一识别。
4、与样本库进行对比，寻求最近似匹配
这步是比较核心的地方，由于要识别的图形每次都是随机变化的，我们不能进行完
全匹配识别，所以使用的是‘欧氏距离’来进行最近似匹配，资料中的《自由手写体
数字识别》里面有详细说明。
（样本库文件是按照匹配的特征通过事先编写的程序进行学习得到的）
该识别思路对目前很多验证码有效，识别速度快，正确率基本还可以（在一定程度内
样本量越大正确率越高），不能识别的情况也不少，比如字符粘连，导致程序无法正确
分割，从而识别失败，有朋友介绍神经网络识别方法不错，有空一定要学习下。-A brief description of ideas:
1, image binarization
Pictures of all the points expressed by 0 or 1, 1 for the effective point, 0 for background. Used here is the largest between-class variance
(Otsu), are introduced in the data.
2, removal of interference points
3, partition
The whole picture is divided into each individual character, in the next step in order to identify 11.
4, compared with the sample database, the search might match the recent
This step is to compare the core places, due to the recognition of each of the graphics changes are random, we can not be finished
Identification of the entire match, so the use of the Euclidean distance to be like the recent match, the information in the "free handwritten
Digital Identification, "which is described in detail.
(Sample library features in accordance with the matching pre-prepared through the process of learning to be)
The identification of a lot of ide Platform: |
Size: 874496 |
Author:yangq |
Hits:
Description: Delphi 7.0下开发 将所有出现在连连看游戏中的图片保存下来. 游戏开始后.逐个与游戏中的图片比较. 将比较结果存成二维数组,计算能连接的点. 发送鼠标点击事件.点击相应的点.
-Developed under Delphi 7.0 will all appear in the picture game Lianliankan preserved. After the start of the game. One by one and the game picture comparison. Will compare the results stored as a two-dimensional array, calculated to the connection point. Sends mouse click events. Click the corresponding point. Platform: |
Size: 139264 |
Author:黄光 |
Hits:
Search - picture compare delphi