FCA as a Tool for Inaccuracy Detection in Content-Based Image Analysis

In this paper we focus on the detection of inaccuracies in the results of content-based image analysis. During the analysis process we detect a set of features, which are later used in Image Retrieval. This detection is based on multiple algorithms specific to particular features. These algorithms use parameters, which have been obtained by the analysis of our test collection. However it seems that in the real application deployment produces some inaccuracies in the results. Our goal is to support the process of feature analysis by detecting these inaccuracies, or at least showing the most probable sources of them. This support can be helpful in tuning these algorithms on less known input data. In the article we describe both the image features detection algorithms as well as usage of Formal Concept Analysis (FCA) as a tool for detection of inaccuracies.