Classification of low-contrast extended objects with application to textile fibers identification

Nowadays there is a demand of intelligent systems which are able to perform classification of color low-contrast extended objects such as textile fibers. Since this is a complex problem with no common way to solve it, detailed investigations are needed here. The paper presents a new way to identify the extended objects such as textile fibers. The workflow consists of three main steps: object extraction, splitting crossed objects and the object identification, and the algorithms for each step are presented. The most complex problem here is the crossed objects splitting which is a subtask of a general clustering task. The paper presents new criteria for the selected object assessment which allows splitting the crossed objects using well-known optimization techniques such as genetic algorithms. Finally, the paper presents the efficiency estimation of the proposed method of the extended objects identification and discusses the possible ways for further enhancement of the mentioned algorithms.