Citation
Khor, Siak Wang and Ahmad, Fatimah
(2008)
Texture-based image search of fashion designs.
In: Knowledge Management International Conference 2008 (KMICe 2008), 10-12 June 2008, Langkawi, Kedah. (pp. 35-37).
Abstract
In today’s world, fashion designers help create billions of dresses, suits, shoes, and other clothing and accessories purchased every year by customers around the world. They study fashion trends, sketch designs of clothing and accessories, select colors and fabrics, and oversee the final production of their designs. With the advent of computers and low storage costs, designers rely on the use of mass computer storage to help them store and manage their designs for future retrievals. The stored designs will be retrieved from time to time so that designers could retrieve them for stimulation of new ideas in their new designs. When the number of stored images grows, searching for a desired piece of design work is likely to be time-consuming and painstaking. Without proper computer-assisted search mechanism, such a search will not be definitely easy. Thus, a huge collection of images containing different designs pattern and textures of certain fashion designs demands for efficient search system. Of all the visual contents identifiable from a design, texture is considered to be the commonest visual attribute that aids in image retrieval. Common approaches for texture-based image retrievals are largely centered around frequency-based models, which are notably suffered from the problem of poor retrieval accuracy. In this paper, a novel approach for texture-based image retrieval has been proposed. This approach, known as Spectral Density Analysis, or SDA for short, is based on the idea of dividing an image into nine equally sized partitions where the spectral density of each partition is computed to be used to aid in the image retrieval process. Benchmarked using the popular performance measurement technique, Recall and Precision, adopting the SDA technique for retrieving images based on their texture contents has produced remarkable retrieval accuracy, almost three times more accurate over any other frequency-based models.
Download File
Additional Metadata
Actions (login required)
|
View Item |