Chapter IX - Image Analysis

The basic concepts of image analysis

Technology to extract information from an image

The main contents

1. Locate the object of interest from the image. Generally, within a scene from the target composition reflects the many areas in the image. In order to separate the area, we want to study image segmentation

2. Since each target different regions or certain features distinguished from other regions apart, edge, texture, shape, color is an important feature. Research on image segmentation and image feature extraction methods can not be separated.

3. The need to use can effectively represent extracted for each target, which requires research shape analysis, representation and description method

4. Analyzing the decision whether there is certain characteristics can be considered a process of identifying, for each target category determination process is also a recognized. High-level analysis of the understanding of the need to use a lot of recognition, pattern recognition is a special course

Characterized in image segmentation

1- task of image segmentation to separate the image into non-overlapping region of interest in order to facilitate further analysis. It is generally the first step in image analysis

2- separate regions generally with respect to the target image, we are interested in

3 important aspects of image processing foundation, are generally more difficult to link

4- other segmentation often rely on the analysis result, accurate segmentation decisions even affect the accuracy of the other part of the analysis

5- segmentation problem is difficult fuzzy interference and noise of the image data

6- So far, there is not a judge is completely correct segmentation criteria, nor is there a standard way to solve all the problems of division. Only some specific issues or required to meet certain conditions of the method

7- segmentation must be judged good or bad from the segmentation results. The actual scene in different situations, analyze specific issues, select the appropriate method according to the actual situation

8- basic characteristic is the gray value image, a color image is its color components. Edge and texture features are also very common feature 

 

Image segmentation

concept:

The image is decomposed into its constituent parts and process objects

The position and scope of the object of interest is selectively positioned in the image

The basic idea

From simple to difficult, step by step segmentation

Control context, reducing the difficulty of splitting

Enhanced focus on the object of interest, the interference reduced the irrelevant image components 

Boundary splitting method

The detection point

If the value of R is equal to 0, indicating that the current detection points around the same point in the gray value

When R is large enough, and the description of the points around the point values very different, it is an isolated point. The controlled T = 32, 64, 128, etc. through the threshold T
| R & lt |> T a detected outlier would

Detection line

边的检测

边界的定义:一段边是两个具有相对不同灰度值特性的区域的边界线

基本思想:计算局部微分算子

适用于:假定问题中的区域是非常类似的,两个区域之间的过渡,仅仅根据灰度的不连续性便可确定

不适用于:当假定不成立时,阈值分割技术一般来说比边缘检测更加实用

 

梯度算子

 

边缘连接法

 由于噪音的原因,边界的特征很少能够被完整地描述,在亮度不一致的地方会中断

因此典型的边界检测算法后面总要跟随着连接过程和其它边界检测过程,用来归整边界
像素,成为有意义的边

Hough 变换

阈值分割法

阈值分割法的基本思想: 确定一个合适的阈值T,将大于等于阈值的像素作为物体或前景,小于阈值的
像素作为背景,生成一个二值图像

全局阈值--对全图采用单一阈值

局部阈值--不同局部采用不同阈值

动态或自适应阈值--每像素点阈值随像素特性而变化

特点:
简单,处理方便

适用于物体与背景有较强对比的情况,重要的是背景或物体的灰度比较单一,确定合适的阈值是关键

这种方法总可以得到封闭且连通区域的边界

噪声、前景或背景灰度变化范围大时将难以应用

通过交互方式得到阈值

通过直方图得到阈值

Isodata门限算法

弓弦法

通过边界特性选择阈值

简单全局阈值分割

分割连通区域

基于多个变量的阈值

面向区域的分割 

数学形态学图像处理

基本概念

形态学:从图像出发,研究数字图像中物体目标的结构及拓扑关系

腐蚀与膨胀

开-闭运算

变体

形态学图像处理应用

边界提取

灰度图像的形态学处理

图像的表示与描述

图像的纹理分析

纹理是图像中一个重要而又难于描述的特征,至今还没有精确的纹理定义;一般地说,纹理是指在图像中反复出现的局部模式和它们的排列规则;或表述为图像强度局部变化的重复模式

纹理图像在局部区域内呈现不规则性,而在整体上表现出某种规律性

一种反映一个区域内象素灰度级空间分布的属性

研究图像的粗糙性、光滑性、规则性等的度量

以纹理特征为主导特性的图像称为纹理图像;以纹理特性为主导特性的区域称为纹理区域。如果区域内部灰度值没有变化,该区域没有纹理

Guess you like

Origin blog.csdn.net/weixin_42572978/article/details/93490542