. Discrimination of images based on their textural features obtained by GLCM
DA analyses were performed in order to investigate the evolution of bread boluses as a function of chewing cycles (C10, C20 and C30). It should be noted that with the TB and RB breads, some subjects swallowed the bolus before 30 chewing cycles (C30) had been completed. Because the number of data corresponding to C30 was lower with these breads, we decided to include images obtained with SW boluses produced after at least 25 chewing cycles in the C30 category(C30∗). This allowed us to compare cycles using a similar number of pictures for each. The Wilks’ Lambda multivariate statistic test indicated a highly significant difference between the chewing cycles for each bread (Table 4). In this Table, “% of correctly classified” indicates for each bread the overall percentage of observations actually belonging to given cycle groups and which are correctly classified in these groups. For example, the use of GLCM textural features for image classification enabled the correct classification of 67.0% of BB images. Classifications per chewing cycle are further detailed in Table 4. “% Classification into” refers to the confusion matrix, indicating the allocation of each observation (image) to a given group. The columns represent the actual groups and the rows the group identified by DA. For example, BB boluses obtained after 10 chewing cycles (C10) are represented by 59 images. Among these images, 62.7% were correctly classified in this group (C10), 35.6% were wrongly classified in C20 and 1.7% were wrongly classified in C30 ( Table 4). Moreover, 73.3% and 65.0% of C20 and C30 images, respectively, were correctly