|M.Sc Thesis||Department of Quality Assurance and Reliability|
|Supervisor:||Dr. Singher Liviu|
The use of machine vision in modern technical systems is spreading fast. One of the problems that may be solved by machine vision is classification of patterns into different groups. The pattern classification problem has a wide variety of applications, for example: identification of targets in real time in military, quality inspections in factories, nondestructive testing, identification of plants of different types used for different agriculture needs, medical diagnoses, archeology and many other applications.
The goal of this work is to design a machine vision system for classification of plant patterns into two different groups in real time. The main problem of plants classification is related to the fact that each plant from one group has its individual and unique form, different than the form of other plants from the same group. Therefore, the plants classification system should have the following properties: high image distortion invariance and noise tolerance. The high speed of image recognition is also a requirement from such system.
Optical pattern recognition with the correlation technique has been used. The VanderLugt correlator has been chosen as the basis of the electro-optical pattern recognition system. A new filtering method based on a discriminant linear programming model, is presented in the work. The method is classifying the light intensity values of each pixel in the output plane of the electro-optical pattern recognition system to two classification regions so that the sum of deviations from the regions’ boundaries is minimized. The method is based on the EMSD (Epsilon Minimize Sum of Deviations) discriminant linear programming model.
The simulation results have indicated good classification ability of the method despite the great shape inconsistency in similar plants. Distortion invariance of the method has been investigated and found to be satisfying for in-plane rotations of an input plant.