A hybrid multivariate classification technique and its application in tissue image analysis

Discriminant analysis and neural network techniques were used in this work to develop a hybrid multivariate classification machine. The design of this machine was based on a theoretical interpretation of human recognition of patterns. This interpretation considers the pattern recognition task as a two-step process: difference detection and classifier training. The classification machine, called machine of classifiers (MC), has two components. The first component performs unsupervised clustering of data based on an iterative Fisher discriminant technique. The second component is a supervised classifier designed on the basis of the committee machine technique. The MC was tested on artificially-generated multivariate data. The results showed the effectiveness of the MC in classification.; Two applications in tissue image analysis were investigated. The first application involved segmenting cartilage and bone areas from other tissues in beef vertebra images. This segmentation was essential for developing an image processing system to determine beef maturity. The MC was applied on image functions in two color systems and successfully segmented the cartilage and bone areas. The second application was meat tenderness prediction. In this application, near-infrared (NIR) images of beef samples were used for tenderness score prediction. The image features were used to develop an MC to predict meat tenderness. The MC was useful in classifying the samples into “tough” and “tender” groups.