The Recognition of the Composition of Paintings by Chinese Flower and Bird Painters

1 I N T R O D U C T I O N In order to produce Expert System for Chinese flower and bird paintings, one needs a data base of the painting composition techniques of the painters. The authors are currently studying the recognition of painters' composition as the first step to making this data base. At this stage, the authors were able to determine the similar painting elements and the composition type in a single painting. Even though, in the recognition experiment, the computer recognition of composition does not perfectly match the arrangement of painting elements in the composition of a painting by an actual painter, the computer is qualitatively able to recognize painting elements and the types of composition.Therefore, this method should be applied for the practical recognition of the painter's style and composition techniques. 2 M A N I P U L A T I O N O B J E C T There are five fixed compositional styles, namely S-shaped composition, triangular-shaped composition, crossed-line-crossing-shaped composition, crossshaped composition and circular-shaped composition [1],[2]. The composition is constructed of painting elements (the basic units which compose a painting, e.g. flowers, leaves, bines, birds, fish and insects). In most cases, the composition selected by a painter belongs to one of the above five composition types. A combination of certain of these composition types may sometimes characterize a particular painter. Thus, one needs to recognize the painting elements and the composition type in paintings of each painter. Paintings by a certain painter can be characterized by the arrangement of the composed painting elements, and thus, the authors are able to reproduce the composition of the painter by recognizing the painting elements and the composition type. In this study, paintings by actual painters were input into a computer using a scanner, and the arrangement of painting elements in the compositions was recognized. Figure l shows some examples of paintings by certain painters. Fig. 1 The i n p u t irnane. 3 RECOGNITION O F T H E A R R A N G E M E N T O F P A I N T I N G E L E M E N T S IN THE C O M P O S I T I O N The composition types (patterns of painting elements arranged in the composition) indicated that for the works of painters, there was a large difference in the patterns. Because of this, when the authors recognized the arrangement by pattern matching, the amount of data increased, because we had to designate the model of composition type pattern for each 4 A MODEL FOR COMPOSITION painting, and the time required for recognition became longer. I t was also difficult to recognize the TYPE PATTERNS A N D COMPOSIarrangement when we input into the computer the paintings of the next painter. Taking these problems TION RULES into consideration, we decided to conduct the recognition process of the arrangement by the following 4.1 MODELS procedures: The models utilized (in the process of recognition) 1. Determine the position of a painting based on the are the composition type model of a single painting image put in the computer. and the model for painting elements each of which composes that painting. Although each model is pro2. Determine where the painting elements are loduced on its own coordinate system, the two are cated. treated conceptually as one composition model as shown in Fig. 2. The following 2 types of parame3. Match painting elements with a single painting ters are given to the model: to determine similar painting elements. 4. Recognize the arrangement by pattern matching using the the model for composition type pattern, because the elements composing a painting are expressed by form and black and white. 5. Determine the composition type based on the composition rules and the relationship between the vertical and horizontal relationships between painting elements. One composition rule is that the conditions including black and white, density and degree of concentration should remain constant [1],[2]. r The parameter whose value changes during the matching process. 1. The ranges of values of the painting elements, sf (f=1,2,3 ,..., n). 2. The value of the relative positions of the frame centroids Gi(XGi, YGi) of painting elements, which compose a painting. r The parameter whose value does not change during the matching process. 1. The coordinate values, g(xgf, ygf), of the position of painting element centroid. 2. The value for the length of the dividing line, T,(i=1,2,3), which is the parameter of the composition type and which was given before the processing, and the value of the ratio of the distance to the tip of the composition line, Tui. 4.2 COMPOSITION RULES a: Twi , Rough position of the t i p s of the divided l ines 0: T u i , Position of the t ips of the cosposition l ines 0: sr, Range of painting elements -: T i , Dividing l ines Comsposition l ines Fig. 2 The composition model of painting element and composition type. Composition rules are used in model matching. Given the composition knowledge, we assume that the range of painting elements and frame centroid will change when we input a painting by a painter into the computer. Thus, we determine a length for dividing lines and determine the position patterns for the tips of composition lines in advance and select some pattern based on the range of painting elements and frame centroid. We gradually change the length of the dividing line based on this pattern and terminate the processing when we decide that the distance to the tips of the composition lines of the model has best matched the value of the ratio. Lr : Radius range of the painting elements +: g(x,r ,y,t ), Centroid of the painting elements F i g . 3(a) The process of painting elements model matching.