Subobject detection through spatial relationships on mobile phones

Reviewer: Charalambos Poullis

This paper presents an "image classification technique for detecting multiple objects (called subobjects) in a single image." Unlike existing techniques, the proposed approach combines image classifiers and "spatial relationships among the subobjects to verify [the object locations] and to predict the locations of detected and undetected subobjects." The approach is then applied and evaluated in the context of a museum guidance system. On the positive side, the application is interesting and relevant. While there are previous examples of approaches to the solutions of this problem, it remains a fairly novel area of application. The approach provides an engaging and intuitive form of information presentation and should be of general interest to the mobile computing and user interface communities, and, to a lesser extent, the computer vision community. The paper is clear and quite well written, with few language errors. The paper has a number of weaknesses, though. The described techniques are relatively simple and straightforward, and seem to be a simple collection of known computer vision methods. Therefore, this paper is a systems/application paper, which is acceptable provided the application is new or the results are particularly outstanding, when compared to other previously published approaches. While the results appear good, the experimentation is, unfortunately, somewhat limited and of average complexity. There is only one real experiment with 12 subobject groups, each containing 3 to 8 subobjects; this is a relatively small and selective sample. In summary, the results are good and the approach is likely correct and promising, but the evaluation is somewhat limited. Online Computing Reviews Service