Abstract:
The use of 1:N identification, which can be easily used with free of hands, for large-scale is expanding. For example, 1:N identification where N=100,000 or more is neede...Show MoreMetadata
Abstract:
The use of 1:N identification, which can be easily used with free of hands, for large-scale is expanding. For example, 1:N identification where N=100,000 or more is needed for large companies or nationwide payment systems. To realize a large-scale 1:N identification, the processing time is a key factor. Multi-step narrowing down is an effective method for speeding up. The multi-step narrowing achieves high speed by combining high-precision features (low speed) and low precision features (high speed), appropriately. However, a major disadvantage of multi-step narrowing down is that it is difficult to find the optimal setting. The possible combinations of features and the narrowing down rate is enormous, and usually the brute-force search was necessary in the past. In this paper, we propose a model that represents the multi-step narrowing down and report the optimal multi-step narrowing down setting based on the model. As experiments, we evaluated the validity of the model using face data. As a result, our proposed model was confirmed valid as an overall tendency. The evaluation of two-step narrowing down shows the optimum total processing time was realized when the processing time of each step was equal or close, as the model predicted.
Published in: 2021 Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC)
Date of Conference: 14-17 December 2021
Date Added to IEEE Xplore: 03 February 2022
ISBN Information:
ISSN Information:
Conference Location: Tokyo, Japan