基于特征提取的图像识别系统的设计首先要解决的问题是从待处理的图像中提取出目标的特征，此特征要求能够很好地区别出目标和背景，使得类间差别最大化，从而降低误检率。 “图代炮”系统即自行反坦克炮瞄准镜视场图像判读模拟系统，其属于图像识别系统的范畴。本文首先分析了待处理图片中的可辨别特征，也即识别目标的大小、形状和相对位置，将这些特征抽象提取出来作为软件系统的设计参考。识别系统的图像处理包括图片的预处理、特征垂线的检测、大立标顶点检测和靶面的匹配。 应用 sobel算子和概率Hough变换可以检测出特征垂线，加快图像处理的速度；使用二值化图像的模板匹配算法对目标进行精确地检测和定位，最后用二值形态学处理检测到的靶面，给出靶面的大小。根据靶面和大立标顶点的相对位置可以判断出命中与否。 6969
关键词：特征提取；图像识别；模拟射击；模板匹配 Title Image recognition system design
based on feature extraction
Abstract
The design of image recognition system is based on the theory of
feature extraction. Extracting the target’s features from the image,
which is to be processed, should be solved immediately and to maximize
the difference in each class and reduce false detecting rate, features
are required to distinguish well between target and background. The
system named TuDaiPao is an image interpretation simulative system
based on self-propelled anti-tank gun’s aiming field and belong to
image recognition system. Firstly, to extract the abstract features
for software design, recognizable features should be analyzed from
the images to be processed, that is recognizing the size, shape and
location of the target. Processing image in the system involves
preprocessing the image, detecting the vertical line and matching the
target with template image. Sobel operator and probability hough
transform can be applied to check out the vertical line by its features
and then accelerate the processing speed. With the template matching
algorithm, the target can be detected and located accurately in the
binary image. The size of target can be detected by using binary
morphology. Given the target and arrow point, we can judge that the
manipulator whether hit the target or not.
Keyword: feature extraction; image recognition; simulative shooting;
template matching 1 绪论 . 1
1.1 应用背景及其意义. 1
1.2 国内外研究概况） 2
1.3 论文结构安排 ） 6
1.4 论文主要工作 ） 6
1.5 本章小结 ） 6
2 特征提取综述 7
2.1 特征提取的基本概念 . 7
2.2 高级特征提取的三种理论 . 7
2.3 特征提取与分类器的区别 . 9
2.4 特征提取的应用） 9
2.5 本章小结 10
3 图像识别系统的要求 ） 11
3.1 一般图像识别系统的工作原理）. 11
3.2 “图代炮”系统的工作原理 ）. 11
3.3 “图代炮”系统的实现要求 ）. 12
3.4 本章小结 13
4 靶标识别技术的原理基础 . 14
4.1 像素间的关系 14
4.2 图像增强技术 15
4.3 图像分割技术 18
4.4 哈夫变换及其改进）. 29
4.5 模板匹配 31
4.6 二值数学形态学 32
4.7 本章小结 35
5 软件架构与实现 ） 36
5.1 库的选择—OpenCV . 36 基于特征提取的图像识别系统设计:http://www.751com.cn/zidonghua/lunwen_4716.html