以影像為基礎之電子商務流行性商品運用系統架構

Abstract

近年來，隨著網路的蓬勃發展，加上成熟的金流與物流環境，電子商務與網路購物已逐漸成為重要的商品銷售通路。網路購物平台如何提供更完善的系統與介面，來協助消費者尋得所需商品，已是電子商務高度競爭中的重要課題。 本論文以協助使用者尋找著重外觀差異的流行性商品為方向，提出了一個整合影像剪裁、內容式影像檢索等既有影像相關技術的系統架構。論文中以此架構為基礎，實作了可萃取與比較尋找鞋類商品影像色彩特性之系統。此系統首先透過JSEG方式將影像進行預先切割，再結合HTML5標準中的畫布元件，透過瀏覽器實現影像剪裁的功能，協助使用者能輕鬆地區分商品影像的前景與背景；我們採用MPEG-7的主要色彩特徵描述來表示及儲存影像色彩特性，以計算效率較佳的LBA演算法與改進的距離函數，用以萃取與比較影像的主要色彩特徵與相似程度，配合主要色彩特徵指定與選取的介面，來協助使用者尋找外觀相近的商品。此系統經過使用者研究，顯示的確可協助使用者更容易尋得所需的商品。

With the tremendous growing of Internet and the well-developed environment of online payment tools and logistic support, the global electronic market has a profound impact on commerce in recent years. Issues of technology usage become critical as businesses and retailers attempt to exploit the boom in electronic marketing. Now more than ever, the promise of online shopping will depend to a great extent upon the interface of online stores. For the attempt to help consumers of online shopping to find fashion products by their appearance, we propose a framework with existing image technology including Image Segmentation and Content-Based Image Retrieval (CBIR). We also implement a system that can find similar shoes based on the color feature of the image. First, we pre-segment the image by JSEG method. We then accomplish the image cutout function on the browser through the Canvas widget of HTML5 standard. This can help user easily separate an image from its background. The MPEG-7 dominant color descriptor (DCD) is chosen to represent and store the image feature. To extract dominant color and determine their similarity, we use linear block algorithm (LBA) and improved distance function which has low computational cost. Finally, user can find similar products through the user interface of our system by selecting sample image or picking dominant color. We have conducted usability studies to compare our system with general e-commerce system. It shows that our system can help users to find products they need more easily.