Gait recognition has received increasing attention for security and authentication since it can be done unintrusively from afar and without a subject's awareness. In this work, we propose a new model-based gait recognition technique called JRC-CNN gait recognition. We introduce three new concepts. (1) We create a new way to preprocess skeleton data by rotating skeleton data using two virtual axes. This process reduces the fluctuation in movements and resolves the multi-viewpoint issue. All postures in a walk are observed from the same angle. (2) We introduce new Joint Replacement Coordinates (JRCs), which represent the movements of the left and right joints in a group of three connected joints. These JRC gait features are designed to put more emphasis on local movements than the movements of non-connected joints. (3) We construct a new Convolution Neural Network (CNN) for the classification process, which consists of a convolutional layer on each JRC and two fully-connected layers. A convolutional layer is designed to discover relations within a group of three connected joints. Fully-connected layers also find the relations of all groups of three connected joints throughout an entire body (in a posture). Our JRC-CNN technique achieves above 98.4% accuracy and significantly outperforms other existing techniques for all free-direction walk datasets. It also performs well under the gallery-size test and the CMC curve test. This means that our proposed JRC-CNN gait recognition technique can be used in a real-world situation. Experimental results also suggest that a person can be identified by a unique posture (an entire body is observed as a whole) with the focus on the movements of connected joints.

With the increase of terrorist threats around the world, human identification research has become a sought after area of research. Unlike standard biometric recognition techniques, gait recognition is a non-intrusive technique. Both data collection and classification processes can be done without a subject's cooperation. In this paper, we propose a new model-based gait recognition technique called postured-based gait recognition. It consists of two elements: posture-based features and posture-based classification. Posture-based features are composed of displacements of all joints between current and adjacent frames and center-of-body (CoB) relative coordinates of all joints, where the coordinates of each joint come from its relative position to four joints: hip-center, hip-left, hip-right, and spine joints, from the front forward. The CoB relative coordinate system is a critical part to handle the different observation angle issue. In posture-based classification, postured-based gait features of all frames are considered. The dominant subject becomes a classification result. The postured-based gait recognition technique outperforms the existing techniques in both fixed direction and freestyle walk scenarios, where turning around and changing directions are involved. This suggests that a set of postures and quick movements are sufficient to identify a person. The proposed technique also performs well under the gallery-size test and the cumulative match characteristic test, which implies that the postured-based gait recognition technique is not gallery-size sensitive and is a good potential tool for forensic and surveillance use.

In order to assist EFL students with low English proficiency in learning writing skill, we propose a framework of a virtual environment to evaluate common errors that the students often conduct in writing essay. The system is a service applied in platform linked with other NLP services to help with language analysis. The system mainly focuses on finding the writing errors related to semantic meaning selection, incorrect structure to indicate the intended meaning, non-smoothing sentence in topic, and etc. The system improves the students' writing skill by providing questions relating to the matters they are writing.

In this paper, we present an innovative approach to integrate spatial relations in stroke clustering for handwritten Devanagari character recognition. It handles strokes of any number and order, writer independently. Learnt strokes are hierarchically agglomerated via Dynamic Time Warping based on their location and their number and stored accordingly. We experimentally validate our concept by showing its ability to improve recognition performance on previously published results.

Inductive learning of first-order theory based on examples has serious bottleneck in the enormous hypothesis search space needed, making existing learning approaches perform poorly when compared to the propositional approach. Moreover, in order to choose the appropiate candidates, all Inductive Logic Programming (ILP) systems only use quantitive information, e.g. number of examples covered and length of rules, which is insufficient for search space having many similar candidates. This paper introduces a novel approach to improve ILP by incorporating the qualitative information into the search heuristics by focusing only on a kind of data where one instance consists of several parts, as well as relations among parts. This approach aims to find the hypothesis describing each class by using both individual and relational characteristics of parts of examples. This kind of data can be found in various domains, especially in representing chemical compound structure. Each compound is composed of atoms as parts, and bonds as relations between two atoms. We apply the proposed approach for discovering rules describing the activity of compounds from their structures from two real-world datasets: mutagenicity in nitroaromatic compounds and dopamine antagonist compounds. The results were compared to the existing method using ten-fold cross validation, and we found that the proposed method significantly produced more accurate results in prediction.

We propose a geometric method for document image processing. This research focuses on document understanding and classification by applying the Winnow algorithm, an online machine learning method. This application makes the document image processing more flexible with various kind of documents since the meaningful knowledge can be extracted from training examples and the model for document type can be updated when there is a new example. This research aims to analyze and classify scientific papers. We conduct the experiments on documents from the proceedings of various conferences to show the performance of the proposed method. The experimental results are compared with the WISDOM++ system and also show the advantages of using the online machine learning method.