Output details

Lost in quantization: Improving particular object retrieval in large scale image databases

A key component in state of the art systems for visual object retrieval from large image datasets is that local regions of images are characterized using high-dimensional descriptors. These descriptors are then vector quantized (VQ) using a discrete codebook. This paper explores methods to represent and match descriptors which were lost in the quantization stage of previous systems. The new methods improve the retrieval performance. This paper has generated a research thread on methods to overcome or avoid the problems of VQ. The methods were used in the spin out company PlinkArt for identifying paintings from a mobile phone,

http://en.wikipedia.org/wiki/PlinkArt