Texture image analysis: application to the classification of bovine muscles from meat slice images

Abstract

Image texture is analyzed to provide a series of features for the classification of several sets of images. Images of meat slices are processed to classify various samples of bovine muscle as a function of three factors: animal age, muscle and castration. The different images present a particular texture that is a global representation of the connective tissue. The aim of texture analysis is to extract specific features for each kind of meat. The meat slices available for this study came from 19 animals, including 10 castrated animals. Their ages were 4 months (10 animals), 12 months (5 animals) and 16 months (4 animals). The same three muscles were studied for each animal. The texture analysis was carried out on digitized images using the first- and second-order statistics of the gray levels and morphological parameters, for the characterization of the marbling. Two classification methods were implemented: the method of the k-nearest neighbors and a method based on neural networks. Both methods give comparable results and lead to satisfactory classification of the samples in relation to the three variation factors. The correlation of the textural features with chemical and mechanical parameters measured on the meat samples is also examined. Regression experiments show that textural features have potential to indicate meat characteristics. © 1999 Society of Photo-Optical Instrumentation Engineers.