Recognition and construction of electronics circuit from paper diagram

Abstract

Symbol recognition in general and handwritten electronic component newlinerecognition is gaining more prominence in the area of Computer Vision. Due to its newlinevast application, large number of research community is actively working in the area newlineof handwritten electronic component recognition. newlineThere are two main challenges that need to be address to be called as effective newlinemethod for handwritten electronic component recognition. (i) Robustness of the newlinemethod, in the sense that the high degree of recognition rate even in the presence of newlinelot of variations to enable the scalability of the solution. (ii) Computational speed, to newlineensure that the solution can be effectively been applied for real time application. newlineDue to importance of the problem and its vast potential practical applications, newlineserval researchers and research communities are actively working towards finding newlinesolutions for the problem of handwritten electronic component and circuit newlinerecognition. Also, several publications can be found in the literature to substantiate newlinethe need of solution for handwritten component and circuit recognition. However, the newlinemain limitation of the existing methods in the area of electronic component and newlinecircuit recognition is that, most of the existing methods either work on very limited newlinenumber of components lacking the scalability or they are extremely slow thereby not newlinesuitable for real time applications. newlineHence, this research work was motivated to address the limitations of the newlineexisting methods to bridge the gaps between the need of the effective solution and the newlineresults of the current state of the art. Through this research we have address, three newlinepotential problems in the area of handwritten electronic component recognition in newlineparticular and handwritten symbol recognition in general as defined below. newlineIn this research work we have proposed a novel algorithm to recognize the newlinetype of circuit using the syntactic pattern recognition thru the application of Finite newlineState Machine. The proposed FSM takes string of symbols as a input and classify the newlinesting into appropriate circuit type. The proposed method is not only faster in terms of newlinecomputation speed and it is also highly robust compared to other existing approaches. newline newlineVIII newline newlineFor effective recognition of circuit type by FSM, it is necessary to convert the newlinehandwritten circuit to its equivalent string representation. In this research work, to newlineconvert handwritten circuit diagram to its equivalent string representation, we have newlineemployed computer vision approach, where the images are first preprocessed using newlineappropriate filtering techniques to remove the noise. Once the images are newlinepreprocessed, further Morphological operations are performed to segment the newlinehandwritten circuit into its constituent parts (Electronic Components). Subsequently, newlineto recognize the type of the segmented electronic components, we have represented newlinethe electronic components in terms of its feature. In this research work we have newlineproposed two approaches for component recognition. The first approach represents newlinethe component using Histogram of Oriented Gradient (HOG) feature and for the newlinepurpose of recognition we have used Support Vector Machine (SVM) classifier. In newlinethe second approach to improve the accuracy of the component recognition, we have newlineused the hybrid featured derived using Uniform Local Binary Patterns (ULBP) and newlineaugmented the ULBP features with Statistical Features. Again for the purpose of newlinerecognition of Electronic Component, we have employed SVM classifier. newlineDatasets plays vital role in development and evaluation of Computer Vision newlineand Machine Learning algorithms. Unfortunately, in the area of electronic circuit and newlinecomponent recognition problem, there is no standard dataset available. To overcome newlinethis problem, thru this research we have created an extensive dataset for both newlinehandwritten components and handwritten circuits. newlineFinally, to establish the performance of the proposed method, we have carried newlineout extensive experiments on large number of handwritten component and circuit newlineimages. The extensive experiments have revealed that the proposed methods are newlinehighly robust and efficient. Also to establish the superiority of the proposed method newlineover existing contemporary algorithms, we have compared the results of the proposed newlinemethods with that of the existing methods. Through the comparative analysis, it has newlinebeen established that the proposed methods outperform most of the existing methods, newlinein terms of speed and accuracy. newline