Online incremental image classification by use of human assisted fuzzy similarity

In this paper we propose a computational scheme for online incremental type classification of images, based on human assisted fuzzy similarity analysis. First of all, two main parameters from each image are extracted in the form of a center-of-gravity and a generalized volume of the image model.. Then their differences for each pair of images are taken as respective features F1 and F2, which serve as inputs of the fuzzy inference system for similarity analysis. This system uses special asymmetrical Gaussian membership functions that are later tuned by using a predefined list of human decisions (similarities). The list includes fixed number of available pairs of images and the objective is to minimize the discrepancy between the human and the computer similarity decision. The proposed online incremental classification scheme starts with an Image Base consisting of several Core Images that are compared with the new sequentially coming images. With a predetermined threshold, the new images are judged as members of an existing class from the Image Base or as new members thus creating a new class that is added to the Image Base. The flexibility and applicability of the proposed human assisted incremental classification is illustrated on an example of 16 flower images and the results are discussed in the paper.