Mobile image search using multi-image queries

Abstract

Visual search has evolved over the years, according to the demand of users. Single image query search systems are inadequate to represent a query object, because they are limited to a single view of the object. Therefore, multi image query search systems have gained importance to increase search performance. We propose a mobile multi-image search system that makes use of local features and bag-of-visual-words (BoVW ) approach. In order to represent the query object better, we combine multiple local features each describing a different aspect of the query image. Employing different features in search improves the performance of the image search system. We also increase the retrieval performance using multi-view query approach together with fusion methods. Using multi-view images provides more comprehensive representation of the query image. We also develop a new multi-view object image database (MVOD), with the aim of evaluating the performance impact of using multi-view database images. Multi-view database images from different views and distances increase the possibility to match the query images to database images. As a result, using multi-view database images increases the precision of our search system. We compare our image search system with a state-of-the-art work in terms of average precision. In our experiments, we use single and multi image queries together with single viewed database. The results show that our image search system performs better with both single and multi image queries. We also performed experiments using MVOD database and show that using a multi-view database increases the precision.