Comparison of object recognition methods on shorthand reports

The paper presents a set of recognition methods with the aim of comparing and selecting the best one. The research is based on a collection of shorthand reports of the XIXth century. The quality of images is low. Collected images were segmented and transferred into a binary form. Altogether, 5823 symbols were received. A control sample consisting of 234 images was selected from the obtained collection in order to test recognition methods. Three methods of image comparison are considered. Advantages and disadvantages of these methods aimed at recognition of shorthand symbols are presented. Based on the results obtained from methods’ analysis, “Shape matching” was proposed as the best recognition method. The similarity measures were computed for all symbols. Subsequent clusterization of the studied collection was performed with the use of computed measures. As a result, the whole collection was divided into 423 classes.