A Study on Image Retrieval and Image Classification Technique for Chinese Painting
Autor: | Chia-Ching Hung, 洪嘉慶 |
---|---|
Rok vydání: | 2008 |
Druh dokumentu: | 學位論文 ; thesis |
Popis: | 96 With the development of the information technology and advancement of digital image processing techniques, an increasing number of digital images are being produced. In the past, owing to insufficient computation speed and storage capacity, storing or processing a large amount of digital images was almost impossible. In nowadays, computer’s processing speed has significantly improved, and the hardware storage capacity has also been expanded. Storing or processing digital images has become not only possible but also popular. However, one issue has arisen with the growing prevalence of digital images. That is, how to fast and accurately retrieve the expected image from a huge database of digital images? Most of the current retrieval techniques are based on keyword search. Users need to manually input all the information of the expected images when performing a keyword search. However, such retrieval method is effort-taking and inefficient, because the returned results may not be the expected images. In recent years, Content-Based Image Retrieval (CBIR) has been extensively researched and discussed. In this method, color histogram is used as an important feature in image retrieval. Nonetheless, the main problem of using color histogram in the retrieval is that color distribution is not considered, so the retrieved images may have similar colors but different color distributions. Precision of image retrieval will be affected as a result. In this study, annular histogram is used. Concentric circles are drawn around the center of image to record the color information. Besides, the method proposed by J. Sun[1] is also modified and applied to Chinese painting images. Texture is another feature for image retrieval [2]. So far, numerous texture-based retrieval methods are available. In this study, a new texture extraction method is proposed. This method converts images into ASCII codes and calculates the statistics of each code to generate a histogram as another feature for image retrieval. Through the promotion of National Digital Archive Program and digitalization of Chinese paintings, more and more artists are now displaying their works online. Common search engines on the Internet provide search only by keywords. To search for a particular Chinese painting, users cannot effectively and accurately obtain the result and may waste a lot of time. CBIR is a relatively more accurate method. Hence, in image retrieval, feature selection is of great importance. Chinese paintings can be classified into various genres. The main characteristics of Chinese paintings are strokes and colors used by artists. In this study, digitalized Chinese paintings are used to construct a database, where paintings are classified by artist. This database contains images of paintings created by five artists [3]. In the experimental chapter, image classification precision and image retrieval precision will be respectively analyzed. In the experiment of image classification precision, color distribution and texture are used as classification features. Textures on the Chinese paintings are converted into ASCII codes to obtain the distribution of textures. Later, color histogram is used to calculate the statistics of each color. Finally, images are classified using Decision Tree. The experimental results show that the precision of the proposed classification method is higher [3]. In the image retrieval experiment, color distribution and texture are used as features for image retrieval. Through feature extraction, similar images in the database are retrieved to achieve search by image. This approach helps users more accurately and conveniently retrieve the expected image using less computational efforts. The experimental results also indicate that the proposed method can effectively enhance the retrieval precision. |
Databáze: | Networked Digital Library of Theses & Dissertations |
Externí odkaz: |