A Study on pattern classification for feature tensor

2014 Fiscal Year Final Research Report

• Project/Area Number: 24700184
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Perception information processing/Intelligent robotics
• Research Institution: National Institute of Advanced Industrial Science and Technology
• Principal Investigator: KOBAYASHI Takumi 独立行政法人産業技術総合研究所, 知能システム研究部門, 主任研究員 (30443188)
• Project Period (FY): 2012-04-01 – 2015-03-31
• Keywords: パターン識別 / 特徴行列 / 特徴テンソル / 部分マッチング

Outline of Final Research Achievements

In this study, we have proposed two novel methods for classifying features represented in a form of matrix or tensor; one is a matrix classifier, and the other is a similarity measure between the feature tensors which is used for exemplar-based classification. The proposed method fast optimizes the matrix classifier by minimizing classification errors as well as a matrix rank to produce a low-rank classifier of high generalization performance. Such low-rank classifier also facilitates to physically interpret the classifier weights for further analysis. The proposed similarity measure is based on partial matching of pair-wise feature tensors. It automatically extracts common patterns shared by those feature tensors and thereby produces effective similarity in disregard of noisy background patterns. In the experiments on various visual recognition tasks, the proposed methods exhibited favorable performance.

Free Research Field

パターン認識