Reducing Noise Interference in Image Query Processing

Query-by-example is the most popular query model for today’s image retrieval systems. A typical query image contains not only relevant objects (e.g., Eiffel Tower), but also irrelevant image areas (e.g., the background). The latter, referred to as noise in this paper, has limited the effectiveness of existing image retrieval systems. To address this drawback, the user must be allowed to draw contours to precisely outline the relevant image areas as the query. We refer this class of queries as noise-free queries (NFQs), and propose a technique for processing them in a sampling-based matching framework. We introduce a new similarity model, and investigate indexing techniques for this new environment. We also implement a prototype to demonstrate this approach. Our query model is more expressive than the standard query-byexample. The user can draw a contour around a number of objects to specify spatial (relative distance) and scaling (relative size) constraints among them, or use separate contours to disassociate these objects. Our experimental results confirm that traditional approaches, such as Local Color Histogram and Correlogram, suffer from noisy queries. In contrast, our method can leverage NFQs to offer significantly better performance. This is achieved using only 1/16 of the storage overhead required by the other two techniques.