Feature Extraction Comparisons Based on Two Different Time-Frequency Analysis

View Statistics Email Alert RSS Feed

Information

Details

Project title

Code/計畫編號

NSC97-2115-M019-002

Translated Name/計畫中文名

兩種不同時頻分析的特徵萃取比較

Project Coordinator/計畫主持人

Mong-Shu Lee

Funding Organization/主管機關

National Science and Technology Council

Department/Unit

Department of Computer Science and Engineering

Website

https://www.grb.gov.tw/search/planDetail?id=1655234

Year

2008

Start date/計畫起

01-08-2008

Expected Completion/計畫迄

31-07-2009

Bugetid/研究經費

452千元

ResearchField/研究領域

數學
資訊科學--軟體

此計畫主要比較經驗模組分解與對偶樹小波分解轉換兩者在手寫數字分類表現，因為對偶樹小波轉換改進過去傳統小波轉換不具平移及旋轉不變性的缺點，因此適合用來表示人為手寫數字的不變性。如果對照以傅立葉為基礎的分解方法，經驗模組分解則是完全資料取向的，即它將數字本身分解成一些與生俱來的基本模組函數，後者這些函數，顯然可視為數字影像的特徵，在應用上我們希望能萃取出數字影像的主要特徵，而忽略其他造成失真的因子，如平移、放大或縮小及旋轉等。初步的實作說明當手寫數字受到平移影響時，如採用經驗模組分解方法則有較好的分類表現，而整體的辨識結果仍有待進一步實作驗證。 This project focuses on the performance comparison between EMD (Empirical Mode Decomposition) and DTWT (Dual-Tree Wavelet Transform) for the classification of handwritten digits. The DTWT improves the shift and rotation invariance on the various traditional wavelet transforms, and suits for representing invariant digit characteristics which depends on writers. In contrast to the standard Fourier based decompositions; EMD is fully data-driven method that decomposes the digit images into their intrinsic mode functions which correspond to the features of the digit images. For our application, we will extract the main feature of the digit and ignore the other distorted factors, such as shifting, scaling and rotation. Preliminary experiments demonstrate that the EMD approach achieves better classification result when the handwritten digits are shifted. The overall classification results still need further investigation.

Keyword(s)

經驗模組分解
小波轉換
特徵萃取
Empirical mode decomposition
Wavelet transform
Featureextraction

DSpace CRIS

Feature Extraction Comparisons Based on Two Different Time-Frequency Analysis

Details

Description