High-accuracy similarity search system

Abstract

The invention provides a high-accuracy similarity search system. A pivot is determined from enrolled data by a pivot determination unit, raw data is acquired, features are extracted from the raw data, a score is calculated as one of a distance and a degree of similarity between the features, an index vector is generated by using the score for the pivot, a score is calculated as one of a distance and a degree of similarity between the index vectors, a parameter of each non-pivot including a regression coefficient is trained by using training data, order to select the non-pivots is, by using the score between search data and the non-pivot as well as the regression coefficient, determined in descending order of posterior probability through logistic regression, and a search result is outputted based on the score between the search data and the enrolled data.

Description

technical field [0001] The present invention relates to methods and systems for retrieving data similar to input unstructured data. Background technique [0002] Retrieving unstructured data similar to input unstructured data such as images, animations, music, documents, binary data, and biological information is called similarity retrieval. Generally, by extracting information called feature quantities used for distance calculation (or similarity calculation) from original unstructured data (hereinafter referred to as raw data), the smaller the distance indicating the degree of inconsistency between feature quantities ( Alternatively, if the degree of similarity indicating the degree of agreement between the feature quantities is large), the similarity is regarded as being more similar, and a similarity search is performed. The distance (or similarity) between feature quantities is called a score. [0003] For example, there is a method of calculating the distance (or sim...

AI Technical Summary

Problems solved by technology

However, in this method, when the calculation of the score of the non-critical data is stopped halfway, regardless of whether the score from the search data is smaller than the threshold value r, the reduction of the non-critical data is terminated without calculating the score (that is, the search is missed). From the point of view of the expected value of the number, that is, there is room for improvement in the point of view of retrieval accuracy

Images