Co-occurrence matrix texture features of multi-spectral images on poultry carcasses

The variance, sum average, sum variance and sum entropy of co-occurrence ma trix were the most significant texture features (probability, P < 0.0005) t o identify unwholesome poultry carcasses at visible and near-infrared wavel engths. When a direction of co-occurrence matrix equals to 0<degrees>, the contrast value was lower and the inverse difference moment and difference v ariance were higher (probability, P < 0.01) than any other direction in the visible spectral images. The characteristics of variance and sum variance of spectral images varied with the wavelength of spectral images and unwhol esomeness of poultry carcasses as well. The sum variance of wholesome was h igher (probability, P < 0.005) than unwholesome carcasses at the wavelength of both 542 and 570 nm. For the near-infrared spectral images at 847 nm, t he sum average, entropy and sum entropy values of unwholesome carcasses wer e higher (probability, P < 0.005) than wholesome ones. The linear discrimin ant model was able to identify unwholesome carcasses with classification ac curacy of 95.6%, while the quadratic model (97.0% accuracy) was better to i dentify wholesome carcasses. (C) 2001 Silsoe Research Institute.