Mohd Sanusi Azmi

Arabic calligraphy identification for Digital Jawi Paleography using triangle blocks

Digital Jawi Paleography is a field of research that helps paleographers to identify authors, origin and date of Jawi manuscripts. This research is important because of the existence of a huge amount of Malay manuscripts with unidentified authors, origin and date. Most researches in the area are for Roman and Hebrew text, whereas researches for Jawi text have just begun recently. In this paper, a novel technique is proposed in order to identify types of Arabic calligraphy in Malay ancient manuscripts that were written in Jawi. The novel technique is based on the triangle blocks that were adapted from scalene triangle. Twenty-one features have been extracted from the triangle blocks.

Arabic calligraphy classification using triangle model for Digital Jawi Paleography analysis

Calligraphy classification of the ancient manuscripts gives useful information to paleographers. Researches on digital paleography using calligraphy are done on the manuscripts to identify unidentified place of origin, number of writers, and the date of ancient manuscripts. Information that are used are features from characters, tangent value and features known as Grey-Level Co-occurrence Matrix (GLCM). For Digital Jawi Paleography, a novel technique is proposed based on the triangle. This technique defines three important coordinates in the image of each character and translates it into triangle geometry form. The features are extracted from the triangle to represent the Jawi (Arabic writing in Malay language) characters. Experiments have been conducted using seven Unsupervised Machine Learning (UML) algorithms and one Supervised Machine Learning (SML). This stage focuses on the accuracy of Arabic calligraphy classification. Hence, the model and test data are Arabic calligraphy letters taken from calligraphy books. The number of model is 711 for the UML and 1019 for the SML. Twelve features are extracted from the formed triangles used.

Digital paleography: Using the digital representation of Jawi manuscripts to support paleographic analysis

Palaeography is the study of ancient handwritten manuscripts to date the age and to localize ancient and medieval scripts. It also deals with analysing the development of the letters shape. Ancient Jawi manuscripts are one of the least studiedarea. Nowadays, over 7789 known Jawi manuscripts are kept in custody of various libraries in Malaysia. Most of these manuscripts were undated with unknown authors and location of origin. Analysing the different types of writing styles and recognizing the manuscript illuminations can discover this important information. In this paper, we discuss the palaeographical analysis from the perspective of computer science and propose a general framework for that. This process involves investigation of Arabic influence on the Jawi manuscript writings, establishing the palaeographical type of the script, and classification of writing styles based on local and global Jawi image features.

Historic Document Image De-noising Using Principal Component Analysis (PCA) and Local Pixel Grouping (LPG)

In this paper, an approach of principal component analysis (PCA) with local pixel grouping (LPG) is used to de-noising the noisy historical document image. This technique ensures the preservation of historic document image local structure. This is due to block matching based LPG which carries out classification to allow only the sample blocks with similar contents used in the calculation for PCA transform estimation. Such an LPG procedure ensures that the image local features can be well preserved after the noise removing process in the PCA domain. The LPG-PCA de-noising procedure will repeat one more times with adaptively adjusted noise level to further improve the performance of de-noising the historic document image. The experiment results show that LPG-PCA model has good results in de-noising historical document image.

Data normalization for triangle features by adapting triangle nature for better classification

Geometry features especially triangle has been widely used in face, fingerprint, vehicle detection and digit recognition. Features from the triangle are used to generate useful features for classification processed. Recently, triangle features used in digit recognition has adopted angle as part of features. This has influenced accuracy due to big gap between angle values and other feature values such as ratio and gradient of sides. To overcome this issue, data normalization can be used to address the issue. Experiments have been made using existing normalization techniques such as Z-score, Minimax and libSVM scale function. Experiments have been conducted using Z-Score and libSVM scale function, but results of classification are worst compared to triangle features without normalization. Thus, the results of classification can be improved by proposed a new technique of normalization based on nature of triangle geometry. In this paper, we have proposed a new normalization technique by adopting the nature of triangle geometry. Datasets HODA, MNIST, IFHCDB and BANGLA digit have been chosen to extract triangle features. Then, we will apply normalization on the extracted features before classify them by using Support Vector Machine. The results shows normalization by adapting the nature of triangle geometry gives better result compared to other techniques. The proposed normalization technique only applies to Cartesian Plane Zone that contributes 45 features. The benchmarking for other researchers should refer to our 25 zones that give 225 features of triangle geometry.

Handling a Small Dataset Problem in Prediction Model by employ Artificial Data Generation Approach: A Review

The emerging era of big data for past few years has led to large and complex data which needed faster and better decision making. However, the small dataset problems still arise in a certain area which causes analysis and decision are hard to make. In order to build a prediction model, a large sample is required as a training sample of the model. Small dataset is insufficient to produce an accurate prediction model. This paper will review an artificial data generation approach as one of the solution to solve the small dataset problem.