Similarity Retrieval for Color Natural Images Based on Multifractal Analysis
碩士 === 國立中興大學 === 資訊科學與工程學系 === 103 === In this paper, we propose a new effective image representation method to capture both color and texture information at the same time from the images for similarity retrieval. First, we convert a color image from RGB to HSV and quantize the hue-component using...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | zh-TW |
| Published: |
2015
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/6xycf3 |
| Summary: | 碩士 === 國立中興大學 === 資訊科學與工程學系 === 103 === In this paper, we propose a new effective image representation method to capture both color and texture information at the same time from the images for similarity retrieval. First, we convert a color image from RGB to HSV and quantize the hue-component using 72 bins. At the same time, the image is also converted into a gray-level image which is subsequently used to calculate the singularity value at every pixel location using multi-fractal analysis method. The hue-component of the image is quantized by 24 bins and the average singularity value of all pixels in the same bin is calculated. Finally, the color histogram and the singularity-value histogram are combined to form a hybrid feature vector of length 96.
We use Corel-1000 dataset to exercise our experiments. There are 1000 natural color images classified into 10 categories in this dataset with each category containing 100 images. We randomly selected 50 images from each category as the query images for our experiments. Then, we compare our method with the MSD method which can also capture color and texture information at the same time in the same structure. The experimental results show that our method outperforms the MSD method in terms of precision and recall.
|
|---|