User preference techniques for support vector machines in content based image retrieval

摘要

Searching multimedia information which allows determining preferences based on very little amounts of data. The preferences are nonparametrically determined. Each preference is quantized into one of a plurality of bins. By doing the quantization, the distances between positive and negative samples are increased. The quantization amount may change depending on the number of samples which are used. The quantization can be used in a support vector machine or the like.