同步旋轉區域三元描述子應用於服飾紋理分類;Synchronized Rotation Local Ternary Pattern for Clothing Texture Categorization

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/74672

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/74672

Title:	同步旋轉區域三元描述子應用於服飾紋理分類;Synchronized Rotation Local Ternary Pattern for Clothing Texture Categorization
Authors:	傅之謙;Fu, Jr-Chien
Contributors:	資訊工程學系
Keywords:	LTP;紋理辨識;旋轉不變性;尺度不變性;流行服飾;LTP;texture classification;rotation invariant;scale invariant;fashion classification
Date:	2017-07-21
Issue Date:	2017-10-27 14:35:47 (UTC+8)
Publisher:	國立中央大學
Abstract:	分析平面影像的資訊是電腦視覺主要的研究領域，其中紋理的識別與分類一直是重要的課題。均勻分布的紋理可以是自然或是人工布料的質地，而在本篇論文裡將研究如何去分類流行服飾上面均勻分布的幾何圖案。這樣一個分類器有助於在影像分割上快速擷取人們的衣著資訊，方便進一步分析服裝款式。傳統的紋理分類方法基本上有LBP、HoG，以及其他變形，近來則是以類神經網路、CNN為主流。雖然CNN具有高準確度，但是相對於傳統的方法，需要大量的資料來去訓練，同時過多的參數需要調配也是缺點之一。本篇論文以改進LTP為主軸進行研究，並以旋轉不變性、尺度不變性為目標將演算法流程加以改良，旨在產生更容易區別分類的特徵向量。在實驗的步驟，共同比較了其他演算法彼此的效能差異，包含面對不同性質的紋理資料集的辨識準確率，以及運算的時間複雜度。而在大部分的情況下，我們的方法都具有不錯的效率與表現。;Texture classification and recognition is an important topic in computer vision research area. In this work, we aim at studying classification of the evenly distributed geometrical pattern printed on the fabrics. This research is helpful to get the clothing information rapidly when performing image segmentation and making the subsequent clothing-style analysis more convenient. Traditional texture analysis features include Local Binary Patterns、Histogram of Oriented Gradients, and other transformation methods. Convolutional neural networks are popular methods in recent year. However, to extract features from CNN, it needs a huge amount of data compared to traditional methods. Also, it is sensitive to parameters. In this paper, we improve LTP and take it as our main work to study. To deal with the problem of rotation invariance and scale invariance, the proposed algorithm generates a feature vector with better discriminability. In experiments, we compare the performance of our method with others on different characteristic texture database. Also, we analyze the time complexity of different algorithms. In general, our method has superior performance and efficiency.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	264	View/Open

社群 sharing

Loading...