The Research Of Muti-domain Sketch Symbol Recognition Based On Semantics

In the field of design,sketch symbol has gained a wide application.People usually record design thoughts through sketch symbol,then further processing through design software.To support user’s sketchy input in design software,the recognition of hand-painted sketch symbol has gained people’s attention widely.The property of low-level independent recognition method of a particular domain’s sketch symbol recognition provides the probability of establishing multi-domain’s sketch symbol recognition.At the same time,correlated domain’s sketch symbol as a semantic unit has semantic information.The domain semantic has great significance to precise the recognition of sketch symbol.Accordingly,the paper proposes a system framework of multiple domain sketch symbol recognition based on semantic Bayesian network,and preliminarily realizes the inference and recognition of sketch semantic symbol based on Bayesian network.The main researches are as followings:(1) Designs a kind of sketch symbol descriptive language named PLE_Sketch which is based on perception.The weight information of human perception embedded in the grammatical structure of PLE_Sketch.PLE_Sketch includes primitives and constraints two parts,which been set initial weight value when establishing.Then,particular domain symbol’s weight value is adjusted through perception filter.The grammatical structure of PLE_Sketch is linear and non-recursive.Limited words can describe unlimited sketch symbols.It is Linear scalability.(2) Presents a domain-independent sketch symbol beautification method and process flow,including main work:first,preprocessing stroke.Then,segmenting the stroke based on corners.Finally,recognizing the primitives and constraints.Proposing a Corner recognition algorithm named GL_TB,which is a Trainable and Backward algorithom based on local and global property of stroke.(3) Proposes sketch semantic symbol inference model based on Bayesian network.According to initial posterior probability,bayes decision generates hypothesis template.The initial posterior probability value of hypothesis template is updated with new stroke joining.According to updated posterior probability,system makes the decision of revocation,hold or generating a larger hypothesis template which contains current hypothesis template. When getting current recognitive result,system generates shape template on the basis of the largest sub-shape segmentation.After filling the empty slot of shape template,system makes error correction management.(4) Completes the design and development of the multi-domain sketch symbol recognition system based on semantics.Besides,three main components of the system are described in details,including multi-domain sketch symbol definition module,domain-independent sketch symbol beautification module and bayes inference module.