A study on feature transformation in pattern recognition

• Project/Area Number: 15K00261
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Perceptual information processing
• Research Institution: National Institute of Advanced Industrial Science and Technology
• Principal Investigator: Kobayashi Takumi 国立研究開発法人産業技術総合研究所, 情報・人間工学領域, 主任研究員 (30443188)
• Project Period (FY): 2015-04-01 – 2019-03-31
• Project Status: Completed (Fiscal Year 2018)
• Budget Amount: ¥4,810,000 (Direct Cost: ¥3,700,000、Indirect Cost: ¥1,110,000); Fiscal Year 2018: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000); Fiscal Year 2017: ¥1,040,000 (Direct Cost: ¥800,000、Indirect Cost: ¥240,000); Fiscal Year 2016: ¥1,560,000 (Direct Cost: ¥1,200,000、Indirect Cost: ¥360,000); Fiscal Year 2015: ¥1,170,000 (Direct Cost: ¥900,000、Indirect Cost: ¥270,000)
• Keywords: 特徴抽出 / 特徴変換 / 画像認識 / 動画像認識 / SVM / パターン識別 / 反転不変 / 固有値問題 / ディリクレ分布 / テンソル / SSIM / ヒストグラム特徴 / ボケ逆変換

Outline of Final Research Achievements

In this study, we have proposed various feature transformation methods to enhance the discriminative power of features. In general, the feature to represent the content of input data contains structural information which is derived from the characteristics of feature extractors and input data distribution. The proposed methods leverage the essential structures to improve the discriminativity of the features. Those methods are formulated especially by focusing on the deblurring of histogram, prior probabilistic models, physical structures and invariance to input data perturbation. We can apply the methods in a computationally efficient manner, while contributing to the improvement of feature representation as well as performance of the whole recognition systems.

Academic Significance and Societal Importance of the Research Achievements

近年、計測データの大規模化・多様化が進み、それらデータをサービス等へ利活用するためのデータ自動認識技術、いわゆるAIの需要が急速に拡大している。本研究成果は、自動認識の中核を成すパターン認識の性能改善に資するものである。特に、特徴抽出の後処理という位置づけで、様々な既存認識システムに容易かつ計算量的にも低コストで導入できるため波及効果も期待できる。さらにそのような実用面のみでなく、既存特徴量の変換処理に着目し、そこに数理的視点を導入した点でも学術的な意義が大きい。

Research Products

• [Presentation] Flip-Invariant Motion Representation (2017)
  • Author(s): 小林匠
  • Organizer: 2017 IEEE International Conference on Computer Vision
  • Int'l Joint Research
• [Presentation] 機械学習を用いた画像特徴量変換 (2017)
  • Author(s): 小林匠
  • Organizer: 第20回情報論的学習理論ワークショップ
  • Invited
• [Presentation] Structured Feature Similarity with Explicit Feature Map (2016)
  • Author(s): Takumi Kobayashi
  • Organizer: IEEE Conference on Computer Vision and Pattern Recognition (CVPR)
  • Place of Presentation: Caesars Palace, Las Vegas, Nevada, USA
  • Year and Date: 2016-06-27
  • Int'l Joint Research
• [Presentation] Histogram Feature Deblurring (2016)
  • Author(s): Takumi Kobayashi
  • Organizer: International Conference on Acoustics, Speech, and Signal Processing (ICASSP)
  • Place of Presentation: Shanghai International Convention Center, Shanghai, China
  • Year and Date: 2016-03-25
  • Int'l Joint Research