THE SEGMENTATION OF THE IMAGES IN THE TASKS OF DETAILED EVALUATION OF THE CONDITION OF MASSIVE QUANTITY OF OBJECTS
Mathematical supplying of the subsystem of segmentation is viewed as decomposition of 2D-image on homo-geneous areas which don’t demand connection of the segments, in the task of identification and evaluation of the condition of sections of areas of massive quantity of solitary material objects (as general aggregate) for quality control of production and acceptance of the solutions during the control of technological process, which adds functionality of the system of computer vision (SCV) of detection of types and sorts of the objects (as elements of general aggregate) on 2D-image with high variability within their types and sorts and near the sorts them-selves. For determination of necessary number of clusters (segments) of the algorithm k-means++ in every 2D-image of an object which is executed by the subsystem, adaptive k-means++ is suggested with iterative deter-mination of necessary number of clusters. The automation of choice and stop on optimal number of clusters uses examination on each step of special condition on reference entropy. The clusterization uses on each step either color coordinates of LAB-pixels or all 5 coordinates of pixels including spatial XY (it means the numbers of line and column of pixels’ matrix). The application only of LAB-coordinates allows to mark out and evaluate the condi-tion of different spots on the surface of objects, the algorithm is sensitive towards small details of the condition of visible surface but without registration of their coherence (the multitude of torn spots with arbitrary outline), it is used for the images with spread segments and is not aimed to search simply connected domains and to mark out the boundary of objects. In the situation of similar color characteristics of different states it is necessary to use the entire LAB+XY coordinates. The efficiency of the subsystem is shown on the images with corn mass of rice, fruits’ raw material and pharmacological production.
Authors: D. S. Ostapov, S. V. Usatikov
Direction: Informatics and Computer Technologies
Keywords: Systems of computer vision, the subsystem of segmentation, adaptive method of k-means++
View full article