Researches On Some Key Technologies Of Computer Calligraphy

With the rapid growth of the digitized calligraphic works and the new requirements of the users, the existed techniques on computer calligraphy cannot meet the needs in digital libraries. There are new challenges in the researches on digital calligraphy:how to retrieve calligraphic characters much faster with a content-based method from a large calligraphy database? As the existed OCR techniques cannot recognize calligraphic characters perfectly while some people want to know the contents of calligraphic works, how can the calligraphic characters be recognized automatically? How can style-consistency calligraphic characters and even calligraphic works be synthesized?To address the above problems, this thesis discusses theories, methods and techniques on computer calligraphy process, which are how to retrieve and recognize calligraphic characters fast, how to extract and recognize the calligraphic strokes, how to quantitatively express the style features in calligraphic characters and how to develop a system for style-consistency calligraphic character synthesis as well as calligraphic tablet synthesis. The main contributions of this thesis are as follows:1. Skeleton extraction for calligraphic characters. As the existed thinning methods cannot extract the accurate skeletons for calligraphic characters, a novel approach called mathematic morphology-fused index table skeletonization (MFITS) is proposed. The calligraphic character is thinned with mathematic morphology firstly and the skeleton is extracted with index table finally.2. Calligraphic character retrieval and recognition based on skeleton similarity. As the contour based calligraphic character retrieval method is slow, a novel approach is proposed to retrieve calligraphic characters with the skeleton similarity. The shape features on the skeletons of calligraphic characters are extracted and analyzed, and the similarity between the shape matrices is evaluated to retrieve the calligraphic characters. As the existed OCR methods cannot recognize calligraphic characters correctly, a novel method for calligraphic character recognition is presented, which evaluating the annotation of the retrieval results in order to recognize the calligraphic character.3. As the calligraphic retrieval is slow while search from a large database, two rapid retrieval methods are proposed, which are the key pixel based filter method and the recognition based retrieval approach. In the key pixels based filter method, a few key pixels are extracted from the contour of the calligraphic character, the similarities br these key pixels are calculated and the similar candidate calligraphic characters are filter. With this method, only 50% of time is needed for calligraphic character retrieval. The recognition based retrieval approach recognize the calligraphic character with a small database firstly, then retrieve the candidate characters from database with the candidate recognition results, and finally calculate the similarities of them and output to the user. With this approach, the accuracy of the retrieval is improved.4. On the area of stroke extraction and recognition for calligraphic characters, a contour corner based approach is proposed for stroke extraction. The corner points are used to segment the contours. The relative segments are then connected and paired to form the strokes. With SVM and rule-based approaches, a novel method to recognize the extracted strokes is proposed, which classifies the strokes with SVM firstly, and then recognizes the special strokes based on some rules.5. As to extract and express the style features of calligraphic characters, a style evaluation model (SEM) is proposed. Both of the general style features and local style features are analyzed. These style features, such as the proportion between height and width, the stroke width, the position and size of each radical, etc, are extracted and the SEM is formed.6. Style-consistency calligraphic character synthesis. A calligraphic character synthesis approach is proposed based on the SEM. A 3-layer model is used to analyze calligraphic characters.With the determination of structures, the proper radicals and strokes are chosen and the new calligraphic character is generated. With the help of SEM, the new calligraphic character is evaluated and adjusted to the particular style.7. Style-consistency calligraphic tablet generation. An approach to generate calligraphic tablet in specific style is proposed. The existed calligraphic characters are retrieved firstly, and the inexistent calligraphic characters are synthesized with the style-consistency calligraphic character synthesis method. At the same time, the layout of the characters is calculated according to the size of the tablet and the number of characters. Finally, the styles of all these calligraphic characters are unified and they are put on the tablet.