Content-Based Image Retrieval System Using Fuzzy Colour and Local Binary Pattern with Apache Lucene

Abstract

The explosive increase and ubiquitous accessibility of visual data on the computer, web, and even smartphones have led to the prosperity of research on image retrieval system. With the ignorance of the visual information and the content of the image in the retrieval process, methods such as text-based image retrieval can lead to inconsistency and inaccuracy between the text search and the image result. A more precise yet powerful technique known as content-based image retrieval (CBIR) can analyse and store the visual information of the image in feature vector representation. However, the big challenge is the semantic gap and intention gap to retrieve relevant images. Numerous CBIR methods have been developed by researchers to identify the best approach. This paper proposed a technique using a combination of fuzzy colour and local binary patterns (LBPs), where the ten bins and 24 bins output from the fuzzy colour system are mapped into LBP histogram. The indexing and searching process utilized Apache Lucene, where the inverted index data structure is applied to boost up the retrieval speed. The proposed method is compared and benchmarked with other techniques such as region-based HSV colour histogram, IOSB SIFT, and traditional fuzzy colour and texture histogram (FCTH). The evaluation is based on the indexing time, searching time, rotation, and scaling invariant, as well as the ability to retrieve mostly similar images. The proposed fuzzy colour and local binary pattern (FCLBP) method passed all the criteria with better accuracy as well as short indexing and searching time.