Image processing and recognition learning Summary

Source: Internet
Author: User

Digital image processing is a technology that analyzes, processes, and processes images to meet visual, psychological, and other requirements. Image processing is an application of Signal Processing in image domains. Currently, most images are stored in digital format. Therefore, image processing usually refers to digital image processing. In addition, the processing method based on optical theory still plays an important role. Digital image processing is a subclass of signal processing. It is also closely related to computer science, artificial intelligence, and other fields. Many traditional one-dimensional signal processing methods and concepts can still be directly applied to image processing, such as noise reduction and quantification. However, an image is a two-dimensional signal, which has its own special side compared with a dimension signal, and the processing method and angle are also different. Most of the concepts used for one-dimensional signal processing have extensions in the two-dimensional image signal field, and some of them become very complex in two-dimensional situations. At the same time, image processing also has some new concepts, such as connectivity and rotation immutability. These concepts have extraordinary significance only in the case of two-dimensional or higher dimensions. Fast Fourier transformation is often used in image processing because it can reduce the data processing capacity and processing time.

Digital image processing is applied in the following aspects:

Photography and Printing)

Satellite Image Processing (Satellite Image Processing)

Medical Image Processing)

Face detection, feature detection, Face Identification)

Microscope Image Processing)

Vehicle barrier Detection)

 

Digital Image Basics

Basic concepts of images, image sampling and quantification, digital image representation, spatial and gray-level resolutions, Image Textures, and some basic relationships between pixels (adjacent pixels, joining, connectivity, areas, and boundaries), Distance Measurement) linear and nonlinear transformations.

Linear transformation: if the transformation function is linear or piecewise linear, this transformation is linear transformation. Increasing the contrast of an image using a linear function improves the quality of the entire image. The effect of increasing the contrast of A (or several) brightness interval in an image by using piecewise linear functions is to improve the quality of the local brightness interval.

Non-linear transformation: When the transformation function is non-linear, it is a non-linear transformation. Frequently used include exponential transformation and logarithm transformation.

RGB (red green blue): Red, green, and blue

CMYK (cyan-magenta-Yellow-black ink): cyan-magenta-Yellow-black

HSI (hue-saturation-intensity): Tone-saturation-intensity

DDB (device-Dependent Bitmap): Device-related bitmap

Dib (device-independent Bitmap): device-independent bitmap

CVBS (composite video broadcast signal): Composite television broadcasting Signal

YUV (also called y Cr CB) is a color encoding method used by European television systems (in PALS ).

 

 

 

 

Digital image storage and display

Image Format

There are two types of images in the computer: vector graphics and bitmapped graphics ). Vector graphs are a series of points, lines, arcs, and other geometric shapes described by mathematical methods, as shown in (. Therefore, the format used to store such images is called the vector image format. The stored data is mainly used to draw the mathematical description of the image. bitmapped graphics is also called raster graphics ), like a TV image, this image is composed of image points (B). Therefore, the format used to store this image is called the bitmap format, which is often referred to as the bitmap format, the stored data is the value that describes pixels.

 

 

Vector and bitmap

Currently, the formats include BMP, GIF, JPEG, MPEG-4, Tiff, PSD,

PNG, SwF, SVG, PCX, DXF, WMF, EMF, lic, EPS, and TGA.

Currently, there are many well-known image processing libraries, such as leadtools, opencv, and leadtools. This is a powerful image multi-media library, but it is charged for registration. Opencv is a cross-platform medium and high-level API. It currently includes more than 300 C functions. It does not depend on other external libraries, although some external libraries can also be used. Opencv is free for non-commercial and commercial purposes. There are also many open-source image libraries, such:

Imagestone, gimp, cximage, etc. Although their functions are not as powerful as leadtools, general image processing can cope with them.

Detailed function Introduction Reference: http://blog.csdn.net/byxdaz/archive/2009/03/09/3972293.aspx

Download opencv source code and documentation: SourceForge. net
Http://sourceforge.net/projects/opencvlibrary/

 

 

Digital Image Enhancement

Image Enhancement aims to improve the image display quality and facilitate Information Extraction and Recognition. In terms of methods, we try to discard information that is deemed unnecessary or interfering, and highlight the required information to facilitate analysis and interpretation or further processing. The following describes several simple remote sensing digital image enhancement methods.

A space domain enhancement processing

The Spatial Field refers to the two-dimensional space of the image plane. The spatial area image enhancement refers to the application of a mathematical model on the image plane to achieve the enhancement by changing the gray level of the image pixel, this enhancement does not change the pixel location. Airspace enhancement includes two types: airspace transformation enhancement and airspace filtering enhancement. Spatial transform enhancement is an enhancement method based on point processing, and spatial filter enhancement is an enhancement method based on neighbor processing.

1) enhanced airspace Transformation

Common methods for enhancing spatial transform include contrast enhancement, Histogram Enhancement, and image arithmetic operations.

Contrast enhancement is an image processing method that changes the contrast of an image pixel by changing the brightness distribution of the image pixel and extending the gray-scale distribution interval to improve the image quality. Because the brightness value reflects the radiation intensity, it is also called radiation enhancement. Common methods include contrast linear transformation and nonlinear transformation. The key is to find a function to transform each pixel in the image, so that the pixel is re-allocated in a unified manner to form an image with enhanced contrast.

Histogram Enhancement

Histogram equalization

The basic practice of histogram equalization is to replace the original random distribution with the probability distribution of each gray interval, that is, the number of pixels in each gray level in the enhanced image is roughly the same. Histogram equalization enhances the details of the largest area, while small areas are merged with those close to gray levels to form a comprehensive area. Reduce the gray level in exchange for increasing the contrast.

Histogram Normalization

Histogram normalization is to convert the histogram of the original image into a histogram of a specified form or a histogram of a reference image, and then adjust the gray level of each pixel of the original image according to the histogram of the specified form known, finally, a Histogram Matching image is obtained. This method is mainly used to match another unsatisfactory image with the histogram of the standard image, to improve the quality of the processed image. For example, when the numbers are embedded, the image tones in the overlapping areas may vary greatly due to equal time. Using Histogram Matching can improve the image tones in the overlapping areas, if the mosaic image has little difference, you can perform seamless mosaic.

Arithmetic Operations on Digital Images

After two or more single-band images are registered in space, image enhancement can be achieved through a series of operations to extract some information or remove some unnecessary information.

 

2) Enhanced airspace Filtering

The enhancement of airspace transformation is calculated point-by-point based on pixels. The image quality is improved as a whole without considering the influence of surrounding pixels. Spatial Filter enhancement aims to highlight certain features of an image (such as highlighting edges or textures). Through the relationship between pixels and adjacent pixels, image enhancement is performed using the neighborhood Processing Method in the spatial domain. Neighborhood processing is used for Noise Removal, image smoothing, sharpening, and related operations.

Image convolution is a local image detection operation in the spatial domain to achieve smooth and sharpening. The specific method is to select a convolution function, also known as "m × n window" or "template", such as 3 × 3 or 5 × 5. Then, an active window of the same size as the template is opened in the upper left corner of the image. The image window is multiplied by the brightness value of the template pixel and then added. Assign the calculation result to the central pixel as its gray value, and then re-calculate the result after moving, assign the calculation result to another central pixel, and so on until the full image scan ends to generate a new image.

Smoothing is used to reduce the variation when the brightness of an image changes too much or when there is an unexpected highlight ("noise, smooth the brightness or remove unnecessary "noise" points. It is actually used to make the image's high-frequency components fade away, that is, to smooth the details of the image, reduce its contrast, and save the low-frequency components, called low-pass filtering in the frequency domain. Specific methods include mean smoothing, median filtering, and sharpening.

Sharpening is used to increase the gray value change rate of the edge to make the border clearer. It is a high-frequency component in the enhancement image. It is called high-pass filtering in the frequency domain processing, that is, to improve the contrast of the image details, also known as edge enhancement. The sharpening method is often used to highlight the edges, linear targets, or parts with a high brightness change rate. There are generally three implementation methods:

(1) Gradient Method

The gradient reflects the brightness change rate of adjacent pixels. If an image has edges, such as boundaries of lakes and rivers, mountains and roads, a large gradient value is displayed at the edge. The brightness gradient value is smaller for the part with a smoother brightness value. Therefore, finding the position where the gradient is large will find the edge, and then replacing the value of the pixel at the edge with different gradient values will highlight the edge and sharpen the image. There are usually the Robert's gradient and souber's gradient methods.

(2) Laplace Algorithm

The Laplace algorithm is different from the gradient method. It does not detect uniform brightness changes, but detects the change rate, which is equivalent to second-order differential. The calculated image highlights the position where the brightness value changes.

(3) Targeted Detection

When purposefully detecting edge, line, or texture features in a certain direction, you can select a specific template convolution operation for targeted detection. It can detect vertical boundary, horizontal boundary, and diagonal boundary. Different templates are used.

 

B. Frequency Domain image enhancement processing
Frequency Domain enhancement refers to the operation of the conversion coefficient (frequency component) of the image in the image's frequency domain. Then, the enhancement effect of the image is obtained through Fourier inverse transformation.

In general, the edge and noise of the image correspond to the high-frequency part in the Fourier transform, so the low-pass filter can smooth the image and remove noise.

The gray fusion part of the image corresponds to the high-frequency component of the spectrum. Therefore, the high-frequency filter can be used to reduce or suppress the low-frequency component and sharpen the image.

The frequency domain is the space that consists of two-dimensional Fourier transformation of the image f (x, y) and the value of the corresponding frequency variable (u, v. The pattern (or pattern) of the image strength in the spatial domain can be directly reflected in the space. F () is the origin in the frequency field, reflecting the average gray level of the image, that is, the DC component in the image. The low frequency reflects the slow gray level of the image; the high frequency corresponds to the area where the gray scale in the image changes more rapidly, such as the edge and noise. However, the frequency domain cannot reflect the spatial information of the image.

 

 

Two-Dimensional DFT and its inverse transformation, fast ft

For more information, see the mathematical knowledge.

Ing between airspace and frequency domain filtering:

Convolution theorem is the most fundamental link between airspace and frequency domain filtering. Two-dimensional convolution theorem:

 

 

 

 

Basic computing process:

Obtain the image of the H (m, n) function about the origin)
After moving H (-M,-N) to a certain distance (x, y), obtain h (x-M, Y-N)
Summation of the product function f (m, n) h (x-M, y-n) in the value range of (m, n)
Displacement is an integer increment, repeating the above process for all (x, y) until two functions: F (m, n) and H (X-M, y-n) there are no overlapping parts.
 

Fourier transform is a bridge between the airspace and the frequency domain. For the Fourier transform pairs of the two domain filters:

 

 

 

 

 

Comparison of frequency domain and airspace Filtering:

1. for spatial and frequency filters of the same size: h (x, y), H (u, v), frequency-domain calculation (due to FFT) it is often more effective (especially when the image size is large ). However, if a small template is used in the airspace to solve the problem, it is often used directly in the airspace.

2. Although the frequency domain filtering is more direct, it is better to use a smaller filter in the airspace. This eliminates the need for Fourier transformation and inverse transformation.

3. Due to more intuition, the design of frequency filter is often used as a guide for the design of airspace filter.

 

Smooth Frequency Domain Filter Type
1. Ideal low-pass filter
2. barworth low-pass filter
3. Gaussian low-pass filter
Frequency Domain sharpening filter type
1. Ideal high-pass filter
2 butworth Qualcomm Filter

3 Gaussian High Pass Filter

4. Laplace operator in the frequency domain
5. Passive templates, high-frequency lifting filtering, and high-frequency reinforcement Filtering
The process of enhancing the frequency domain image processing:

 

 

Image Restoration
Image Restoration: attempts to use the prior knowledge of the degradation process to restore the original face of the degraded image.

 

Basic Idea of Image Restoration: first build a degraded mathematical model and then fit the degraded image based on the model.

Image Restoration models can be processed using continuous mathematics and discrete mathematics. The implementation of processing items can be convolution in the spatial domain or multiplied in the frequency domain.
References:
Http://download.csdn.net/source/1513324

 

Edge Detection

Edge Detection of digital images is an important foundation for image analysis, such as image segmentation, target area recognition, and region shape extraction. The first step in image understanding and analysis is usually edge detection, currently, it has become one of the most active topics in the field of Machine Vision and plays an important role in engineering applications. Edge is the most significant part of the image's local brightness changes. It is the most basic operation for detecting the significant changes in the image's local brightness. Edge records include two types of chaincodes and line segments. The chaincodes are suitable for perimeter calculation. line segments are easy to calculate area and related areas. They can be converted to each other.

Common Edge Detection Algorithms:

Berts Edge Detection Operator

Sobel edge operator

Prewitt edge operator

Kirsch Edge Operator

Canny edge detection

 

Image Compression
Image compression is an application of Data Compression Technology in digital images. It aims to reduce redundant information in image data and store and transmit data in a more efficient format. Image compression can be lossy or lossless. Lossless compression is preferred for technical charts, charts, or cartoons, because of lossy compression methods, especially when low bit rates. For example, for medical images or scanned images for archiving, The lossless compression method is also recommended. Lossy methods are very suitable for natural images. For example, small loss of images in some applications is acceptable (sometimes imperceptible), which can greatly reduce the bit rate.

Lossless Image Compression methods include:

Travel Length Code

Entropy Coding

LZW algorithm

Lossy compression methods include:

Reduce the color to the commonly used color in the image. The selected color is defined in the color palette of the compressed image header. each pixel in the image is represented by a color index in the color palette. This method can be used with the EN: dithering to blur the color boundary.

Color sampling, which uses the human eye's sensitivity to brightness changes much greater than color changes, can reduce the color information in the image by half or more.

Transform encoding, which is the most common method. First, Fourier correlation transformations such as discrete cosine transform (DCT) or wavelet transform are used to quantify and Compress With Entropy encoding.

Fractal Compression ).

 

Morphological Image Processing
Expansion and Corrosion

Expansion
Corrosion
Open and closed operations
Change when hit or not hit
Some basic morphological Algorithms

Boundary Extraction
Area Filling
Extraction of connected components
Convex Shell
Details
Roughening
Skeleton

Crop

 

Image Segmentation
Image segmentation refers to some method to divide the objects in the image scenario into different categories. Generally, image segmentation is implemented by dividing images into two categories: "black" and "white", which represent two different objects respectively.

Image segmentation methods: threshold, region, mathematical morphology, and Pattern Recognition

A. Threshold Value Segmentation includes the following types:

(1) Select the threshold value from the histogram gray distribution

(2) Select the threshold value using the Double Peak Method

(3) iterative method selection threshold

The principle is as follows, which is easy to understand.

Iteration is based on approximation. The steps are as follows:
1. obtain the maximum and minimum gray values of the image, which are recorded as zmax and zmin respectively, and set the initial threshold T0 = (zmax + zmin)/2;
2. Based on the threshold value TK, the image is divided into foreground and background, and the average gray scale values Zo and ZB of the two are obtained respectively;
3. Obtain the New Threshold Value TK + 1 = (Zo + zb)/2;
4. If TK = TK + 1, the obtained value is the threshold value. Otherwise, the value is converted to 2 for iterative calculation.

(4) Dajin method selection threshold

The Dajin method is the most inter-class variance method. It is a simple and efficient method for adaptive Calculation of a single threshold, or Otsu)

The Dajin method was proposed by Dajin in 1979. For image, note T as the segmentation threshold between foreground and background. The ratio of foreground points to image points is w0, and the average gray scale is U0; the proportion of the number of background points to the image is W1, and the average gray scale is U1. The average gray scale of the image is u = W0 * U0 + W1 * U1. From the minimum gray value to the maximum gray value traversing T, when T makes the value G = W0 * (u0-u) 2 + W1 * (u1-u) 2 maximum t is the best threshold for segmentation. The Dajin method can be understood as follows: This formula is actually an inter-Class Variance. The foreground and background produced by the threshold t constitute the entire image, and the foreground value U0 and the probability is W0, the value of the background is U1, the probability is W1, and the total mean is U. This formula is obtained based on the variance definition. Because variance is a measure of gray distribution uniformity, the greater the difference, the greater the difference between the two parts of the image, when some target errors are classified into the background or some background errors are classified into the target, the difference between the two parts is reduced. Therefore, the maximum variance division between classes means that the error score probability is minimized. The calculation amount of the direct application of Dajin method is large, so the equivalent formula G = W0 * W1 * (u0-u1) 2 is generally used.

(5) Select the threshold from the grayscale stretch

Dajin method is a common method, but it will lose some overall information when the gray scale of the two groups of objects is not obvious. So in order to solve this problem, the grayscale tensile reinforcement Dajin method is used. Increase the gray level in Dajin's thinking to enhance the gray level difference between the first two groups of objects. The original gray level is multiplied by the same coefficient, thus expanding the gray level of the image. The test results show that the split effect varies greatly with different tensile coefficients.

 

B. Division of Regions

Regional Growth, regional separation and merger

Region Growth algorithm

C. segmentation based on morphological Watershed

Watershed segmentation algorithm

 

Image Feature Extraction and matching

Common image features include color features, texture features, shape features, and spatial relationship features.

A color features

Features: A color feature is a global feature that describes the surface properties of a scene corresponding to an image or an image area. Generally, color features are based on Pixel features. At this time, all pixels belonging to the image or image area have their respective contributions. Because the color is not sensitive to changes in the direction and size of the image or image area, the color features cannot well capture the local features of objects in the image.

Common Feature Extraction and matching methods:

Color histogram

Its advantage is that it can briefly describe the global distribution of colors in an image, that is, the proportion of different colors in the entire image, it is particularly suitable for describing images that are difficult to automatically split and images that do not need to consider the spatial location of objects. Its disadvantage is that it cannot describe the local distribution of colors in the image and the spatial location of each color, that is, it cannot describe a specific object or object in the image.

Color histogram feature matching methods: histogram intersection, distance, center distance, reference color table, and accumulative color histogram.

 

B texture features

The texture feature extraction method is simple. It uses an active window to slide continuously on the image, calculate the variance, mean, maximum, and minimum values in the window, and their difference and information entropy respectively,

When the spectral characteristics of the target are close to each other, the texture features play a positive role in distinguishing the target. Select an appropriate range of dynamic data changes. After texture features are extracted, the texture features of the image are highlighted, which is conducive to the extraction of structural information.

Features: texture features are also global features, which also describe the surface properties of the scene corresponding to the image or image area. However, because texture is only a feature of the surface of an object and cannot fully reflect the essential properties of an object, it is impossible to obtain high-level image content only by using texture features. Unlike color features, texture features are not pixel-based features. They need to be calculated statistically in areas containing multiple pixels. In pattern matching, this regional feature has great superiority and won't be able to match successfully due to local deviation. As a statistical feature, texture features often have rotation immutability and have strong resistance to noise. However, texture features also have their disadvantages. One obvious drawback is that when the image resolution changes, the calculated texture may have a large deviation. In addition, due to the potential impact of illumination and reflection, the texture reflected in the 2-D image is not necessarily the real texture of the surface of the 3-D object.

Common Feature Extraction and matching methods:

Classification of texture feature description methods

(1) A typical statistical method is a texture feature analysis method called the gray-scale symbiotic matrix. gotlieb and kreyszig, based on the study of various statistical features in the symbiotic matrix, through experiments, the four key features of the gray-scale symbiosis matrix are obtained: energy, inertia, entropy, and correlation. Another typical method in the statistical method is to extract texture features from the image's self-correlation function (that is, the image's energy spectrum function), that is, through the calculation of the image's energy spectrum function, extracting features such as texture fineness and directionality

(2) Geometric Method

The so-called geometric method is a texture feature analysis method based on the texture elements (basic texture elements) theory. According to the texture primitive theory, a complex texture can be composed of several simple texture elements which are arranged in regular order. There are two kinds of algorithms that affect the comparison in the geometric method: voronio board feature method and structure method.

(3) Model Method

The model method is based on the Image Construction Model and uses the model parameters as texture features. A typical method is the random field model method, such as the Markov (Markov) Random Field (MRF) model method and the Gaussian random field model method.

(4) Signal Processing

Texture Feature Extraction and matching mainly include: gray-level co-occurrence matrix, Tamura texture features, self-regression texture model, and wavelet transformation.

The Feature Extraction and matching of the gray level co-occurrence matrix mainly depends on four parameters: energy, inertia, entropy, and correlation. Based on Human Visual Perception psychology research on texture, Tamura texture features propose six attributes:

: Roughness, contrast, direction, line image, normalization, and rough. Simultaneous Auto-regressive (SAR) is an application example of the MRF model.

 

C shape features

Features: Various search methods based on shape features can effectively search for objects of interest in the image, but they also have some common problems,

Common Feature Extraction and matching methods:

Generally, shape features have two types of Representation Methods: contour features and regional features. The contour feature of an image mainly targets the outer boundary of an object, while the regional feature of an image is related to the entire shape area.

Several typical methods for describing shape features:

(1) Boundary feature method this method obtains image shape parameters by describing boundary features. Among them, the methods for detecting parallel lines and the histogram of the boundary direction are classic methods. Using the global features of the image, we can connect edge pixels to form a region Closed Boundary. The basic idea is the parity of point-line; the boundary direction histogram method first obtains the edge of the image from a differential image, and then generates a histogram of the edge size and direction. The general method is to construct the Gray Gradient Direction matrix of the image.

(2) Fourier shape descriptor Method

The basic idea of Fourier shape descriptor is to use the Fourier transform of the Object Boundary as the shape description, and convert the two-dimensional problem into a one-dimensional problem by using the closeness and periodicity of the region boundary.

Three shape expressions are derived from the boundary points, namely the curvature function, the center distance, and the complex coordinate function.

(3) geometric parameter method

Shape expression and matching use a simpler method of regional feature description, for example, a shape factor (such as a moment, area, perimeter, etc.) related to shape quantitative measurement ). In the QBIC system, geometric parameters such as roundness, eccentric heart rate, spindle direction, and algebraic moment are used to search images based on shape features.

 

D. Spatial Relationship features

Features: the so-called spatial relationship refers to the spatial location or relative direction relationship between multiple targets separated from the image, these relationships can also be divided into connection/adjacent relationship, overlapping/overlapping relationship, and inclusion/inclusive relationship. Space Location Information can be divided into two types: relative space location information and absolute space location information. The previous relationship emphasizes the relative situation between targets, such as the upper-lower-left relationship. The latter relationship emphasizes the distance and azimuth between targets. Obviously, relative spatial locations can be introduced from absolute spatial locations, but it is often relatively simple to express relative spatial location information.
The use of spatial relationship features can enhance the ability to differentiate the image content, but spatial relationship features are often sensitive to image or target rotation, reversal, scale changes, and so on. In addition, in practical applications, it is often not enough to use only spatial information, and scene information cannot be effectively and accurately expressed. In order to search, in addition to using spatial link features, other features are also required.

Common Feature Extraction and matching methods:

There are two ways to extract spatial relationship features: one is to automatically split the image and divide the objects or color areas contained in the image, then, the image features are extracted based on these regions and indexed. The other method is to divide the image evenly into several rule sub-blocks, and then extract features from each image sub-block, and create an index.

 

 

Pattern Recognition

Pattern recognition is based on a large amount of information and data, the shape, pattern, curve, number, character format, and graphics are automatically recognized by computer and mathematical reasoning. Pattern recognition involves two interrelated stages: the learning stage and the actual stage. The former selects features of samples and searches for classification rules, the latter classifies and recognizes unknown sample sets based on classification rules. Pattern Recognition in the broad sense belongs to the field of Intelligent Simulation in computer science and has a wide range of contents, including voice and language recognition, text recognition, fingerprint recognition, sonar and seismic signal analysis, photo and image analysis, and chemical pattern recognition. Computer Pattern Recognition automates some mental work.

Pattern Recognition: process and analyze various forms of information that characterize things or phenomena (numerical values, text and logical relationships, the process of describing, identifying, classifying, and interpreting things or phenomena is an important part of information science and artificial intelligence.

The mode can also be divided into two forms: abstract and concrete. The former, such as consciousness, thoughts, and arguments, belongs to the scope of concept Recognition Research and is another branch of AI research. We refer to pattern recognition mainly for speech waveforms, seismic waves, ECG, EEG, pictures, texts, symbols, three objects and scenes, as well as various physical, chemical, and biological sensors. classifies and identifies the specific pattern of an object for measurement.

Pattern Recognition refers to the classification and description of a series of processes or events. Some similar processes or events are classified into one type. The following four methods can be used to analyze and handle pattern recognition problems.

Statistical Pattern Recognition: Statistical Pattern recognition is a recognition method inspired by the decision-making theory in mathematics, it generally assumes that the identified object or feature extraction vector is a random variable that conforms to a certain distribution rule. The basic idea is to define the feature vectors obtained in the feature extraction stage in a feature space, which contains all the feature vectors and different feature vectors, or different types of objects correspond to a point in space. In the classification stage, the feature space is divided based on the statistical decision principle to identify objects with different features. In statistical pattern recognition, the statistical decision classification theory is relatively mature, and the focus of research is feature extraction. The basic principle of statistical pattern recognition is that samples with similarity are close to each other in the pattern space and form a "group", that is, "together by things ". The analysis method is based on the feature vectors xi = (xi1, xi2 ,..., Xid) T (I = 1, 2 ,..., N), classify a given pattern into Class C ω 1, ω 2 ,..., In ω C, the classification is determined based on the distance function between the modes. T indicates transpose, N indicates the number of sample points, and D indicates the number of sample features.

The main methods for Statistical Pattern Recognition include: discriminant function, k-nn, nonlinear ing, feature analysis, and main factor analysis. (Shenmeshi.cn collection)

In statistical pattern recognition, Bayesian decision-making rules theoretically solve the problem of optimal classifier design. However, the implementation of Bayesian decision-making rules must first solve the more difficult issue of probability density estimation. BP Neural Networks learn from observation data (training samples) directly, which is a simpler and more effective method and has been widely used. However, it is a heuristic technique, lack of a solid theoretical foundation for specified engineering practices. The breakthrough achievements made by the Institute of statistical inference theory lead to the establishment of the modern statistical learning theory-VC Theory. This theory not only answered the theoretical problems in the artificial neural network on a strict mathematical basis, in addition, a new learning method, support vector machine, is derived.

 

Artificial Neural Network Pattern Recognition: The study of Artificial Neural Networks originated from the study of biological neural systems. The biggest difference between artificial neural networks and other recognition methods is that they do not need to have too much analysis and understanding on the objects to be identified, and are capable of Intelligent Processing.

Syntactic Structure Pattern Recognition: Also known as the structure method or linguistic method. The basic idea is to describe a mode as a simple combination of sub-modes. The sub-mode can also be described as a simpler combination of sub-modes, and finally get a tree structure description, the simplest sub-mode at the underlying layer is called a schema primitive. Selecting a primitive in the syntax method is equivalent to selecting a feature in the decision theory method. It is usually required that the selected primitive can provide a compact description of the schema to reflect its structural relationship, and be easily extracted using non-syntactic methods. Obviously, the primitive itself should not contain important structure information. A mode is described by a group of elements and their combinations. It is called a mode description statement. This is equivalent to a combination of sentences and phrases in a language, with the same combination of words and characters. A rule that combines elements into a pattern, which is specified by the so-called syntax. Once the primitive is identified, the recognition process can be performed through syntactic analysis, that is, to analyze whether the given schema statement conforms to the specified syntax. If a syntax is satisfied, it is classified into the class.

Among several algorithms, statistical pattern recognition is the most classic classification recognition method, which is widely used in image pattern recognition.

 

 

Reference books: Version 2 of Digital Image Processing in the United States

 

This article from the csdn blog, reproduced please indicate the source: http://blog.csdn.net/byxdaz/archive/2009/07/23/4375228.aspx

Contact Us

The content source of this page is from Internet, which doesn't represent Alibaba Cloud's opinion; products and services mentioned on that page don't have any relationship with Alibaba Cloud. If the content of the page makes you feel confusing, please write us an email, we will handle the problem within 5 days after receiving your email.

If you find any instances of plagiarism from the community, please send an email to: info-contact@alibabacloud.com and provide relevant evidence. A staff member will contact you within 5 working days.

A Free Trial That Lets You Build Big!

Start building with 50+ products and up to 12 months usage for Elastic Compute Service

  • Sales Support

    1 on 1 presale consultation

  • After-Sales Support

    24/7 Technical Support 6 Free Tickets per Quarter Faster Response

  • Alibaba Cloud offers highly flexible support services tailored to meet your exact needs.