Frontiers in Pattern Recognition and Artificial Intelligence

The following sections are included:

• Preface
• Contents
• List of Contributors

The smart devices that we have become so familiar with—and dependent on—make use of three new technologies: miniaturized computers; the internet; and Artificial Intelligence. In this overview, we provide a definition and a brief history of Artificial Intelligence; mention some of the key developments in formal reasoning, natural language processing, robotics, and games playing; and outline the dominant software technologies in use today. We conclude with a discussion of trends, including the possibility of Artificial General Intelligence.

The International Conference on Artificial Intelligence and Pattern Recognition (ICPRAI 2018) brought together scholars to celebrate works on artificial intelligence and pattern recognition. In acknowledgment that two terms are closely related and are sometimes used interchangeably, a Competition was set up as part of the conference whereby attendees were invited to submit a response to the following question: What is the relation between Artificial Intelligence and Pattern Recognition? This chapter includes the top three responses as judged by the ICPRAI 2018 Competition Committee. Barz argues that intelligence goes far beyond pattern recognition, while pattern recognition is an essential prerequisite for any intelligence, artificial or not. Mukherjee and Lall outlines the inter-related concepts in AI and PR and various potential use-cases. Lastly, Elham provides her perspective, which includes seeing pattern recognition as a starting point or a principle of discovery of artificial intelligence developments.

This chapter provides an overview of the field of Natural Language Processing (NLP), a sub-field of Artificial Intelligence (AI) that aims to build automatic systems that can understand or produce texts in natural language. The intended audience is a non-technical reader with no particular background in linguistics or computer science.

The chapter first characterizes natural language and explains why dealing with such unstructured data automatically is a challenge. Examples of typical applications of NLP are then provided ranging from low-level tasks to end-user everyday systems. As much of the work in AI, NLP has gone through three main eras: symbolic approaches, machine learning driven approaches, and more recently, deep learning driven approaches. These three paradigms will be described, with a particular emphasis on the current one, deep learning, which, in only a few years, has led to exciting results and allowed applications, such as conversational agents and machine translation, to become accessible and usable to the public.

A great number of handwritten documents have been digitized, to preserve, analyze, and disseminate them. These documents are of different categories, being drawn from fields as diverse as history, commerce, finance, and medicine. Developing efficient means of analyzing and recognizing these documents is of significant interest. Meanwhile, many devices such as smart phones generate handwritten samples, which need to be recognized and translated to machine code. Therefore. handwriting recognition is an active research area with an increasing number of applications on handwriting. This chapter introduces handwriting recognition, describes the steps of a handwriting recognition process, and finally, summarizes different handwriting recognition applications.

In computational pedagogy, a relatively simple Optical Character Recognizer system can robustly evaluate objective response types automatically without human intervention. This saves time, cost and man-hours. Thus, the next question becomes whether it is possible to develop an automated system for evaluating descriptive handwritten answer types. Recent experiences show that human evaluation of long examination responses is quite subjective and prone to challenges like inexperience, negligence, lack of uniformity in the case of several evaluators, etc. In this work, we present the roadmap for an automated vision system that evaluates descriptive answers based on extracting relevant words and finding the relationship between words according to weights. We introduce context features to handle variations of words written by different users to estimate the final score.

A novel method for Chinese writer identification is proposed in this paper, which takes advantage of both text-independent and text-dependent characteristics. The contour-directional features were extracted from a whole image. They were used to calculate the text-independent similarity between the query and reference handwriting images. Meanwhile, character pairs, appearing in both the query and reference handwriting images, were utilized for computing the text-dependent similarity. We propose an effective method to measure the similarity of character pairs. It is rooted from image registration. The displacement field used to align two characters was calculated by the Log-Demons algorithm, and was utilized for a similarity measurement. The final similarity between the query and reference handwriting images is the fusion of text-independent and text-dependent similarities. The best Top-1 accuracy on the HIT-MW and CASIA-2.1 datasets reached 97.1% and 98.3% respectively, which outperformed other previous approaches.

The relation between handwriting and Alzheimer’s Disease (AD) as well as Parkinson’s Disease (PD) has been studied so far. However, it is just in the last decade that Computer Aided Diagnosis system have been investigated by the scientific community. This work intends to provide a hands-on description from a pattern recognition perspective: state of the art on online data acquisition, feature extraction and classification is described. Successively experimental evidences are provided on two datasets.

This work takes place within the ACTIF project, in the context of the eFran call for projects that aims for active and collaborative learning promotion. This paper presents a pattern recognition and analysis system for Geometry learning in middle school. The goal is to allow students to draw geometric shapes on a pen-based tablet, given a teacher’s instruction. To make the student active, the system have to recognize and analyze on the fly the hand-drawn student’s productions in order to produce real-time visual, corrective, and guidance feedback. We base our work on the visual grammar CD-CMG (Context Driven Constraints Multi-set Grammar), to model the domain prior knowledge and interpret the hand-drawn sketches on the fly. Our first contribution lies in adapting this grammar to the Geometry domain to cover the geometric objects taught in middle school curriculum. Although being expressive enough to model this large scope, the original formalism could not cope with the requirement of real-time analysis, given that the multiple interactions between geometric objects generate combinatorial issues. Our second contribution lies in extending the formalism to obtain acceptable performance for a real-time user interaction system. The first experiments on complex geometric figures drawing scenarios show that the proposed approach allows complexity and interpretation time reduction. We present also our result on another application domain, architecture plan sketching, to prove the generecity of our approach.

Detection of counterfeit coins is a meaningful yet challenging topic due to their small physical size and quick improving forging quality. In this paper, we present a visual measurement methodology on automatic counterfeit coin detection. This system consists of lettering extraction, lettering matching, feature measurement and counterfeit detection. For validation, we tested 3 sets of samples, and the results demonstrate almost perfect accuracy.

This chapter proposes a robust counterfeit coin detection method to tackle the advancements and sophistications of modern forgery methods. The proposed method uses transfer learning by fine-tuning a pre-trained convolutional neural network (CNN) and then analyzes the features and measures of characters on the coin surface. The fine-tuning process customizes the general image features from the original images used for training the CNN i.e. natural image into modified features that are suitable for coins. On the other hand, characters represent one of the two major parts of the coin stamp and is the one used mainly by human vision system to recognize and authenticate coins. The ensemble method combines the classification results of two classifiers trained by features from convolutional layers and a third classifier trained on character features e.g. distances between characters, stroke width, height and depth of characters. The ensemble method achieved a precision rate as high as 85.1% demonstrating the reliability of combining fine-tuned CNN classification results and the classification results of another classifier based on hand crafted features from characters to authenticate coins. The method is evaluated on a real life dataset of Danish coins as part of collaborative research with Danish authorities.

Cardiac murmurs are the first signs of pathological changes in heart valves. Their subtle presence poses great challenges for detection through auscultation or phonocardiograms (PCGs); therefore computer-aided detection (CAD) of heart murmurs has medical significance in assisting health care professionals. Traditional CAD approaches, relying on engineered features, are prone to changes in environmental noise and data collection methods. Deep Convolutional Neural Networks (CNN) have shown robustness in advancing the performance of computer vision tasks through automatic feature learning from large amount of data. Meanwhile, deep Recurrent Neural networks (RNN) have demonstrated state-of-the-art performance on processing sequence data in areas such as speech recognition and natural language processing. With a limited set of labelled PCG recordings, this work first transforms PCGs to topology-preserving spectral images, conducts data augmentation and successfully tunes latest deep CNN architectures with transfer learning. Experimental results indicate that the fine-tuned deep CNNs are effective in classifying cardiac murmurs from PCG recordings without segmentation. Moreover, the deep CNNs gained a further performance boost from learning the dynamics in temporal sequences after being plugged into an RNN model. In summary, the proposed deep recurrent-convolutional neural network approach captured spectral, temporal and spatial features. It achieved an overall accuracy of 94.01% for automatic cardiac murmur detections in phonocardiograms.

Many automobile repair shops and parking lots are often equipped with a sensing coil to detect when a vehicle enters. However, the single sensing coil often produces false alarms (e.g., when motorcycles pass over the sensor or vehicles reverse from or stop at the gates). To overcome these problems, surveillance camera equipment is utilized and installed near the entrance gate to detect the incoming vehicles. Given that image-processing technology is now more developed, we propose using a camera to perform the task of a sensor and improve the accuracy of vehicle detection. This system is empowered such that it can adapt to the different factors of an outdoor environment. The background subtraction method is utilized to extract foreground information of a moving object from the field of view of the camera. Optical flow is also applied to determine the direction of the incoming vehicle. A foreground energy-map analysis is conducted to improve the reliability of vehicle detection. Finally, a 20-day test of the system was conducted at an automobile repair shop. The experimental results show a highly promising detection performance with a hit rate as high as 95.58%. The camera system is designed to replace the sensing coils to reduce the hardware costs and increase the detection efficiency.

Compressive tracking algorithm is a simple and efficient tracking algorithm, which has good accuracy and a robust performance. Despite much success that has been demonstrated, numerous issues remain to be addressed. First, the samples selected by the algorithm do not have discriminative representation and the number of samples is relatively huge affecting the computational complexity. Furthermore, the search window of the algorithm is fixed, so when the target scale changes, noise is introduced into the Bayes classification model easily and leads to the loss of tracking. In this chapter an automatic compressive tracking based on motion prediction is proposed. The number of samples is reduced and the tracking speed is improved by using motion prediction, the tracking accuracy is also improved by adjusting the tracking window size adaptively. Extensive experiments in subjective and objective comparison are performed on the OTB datasets and the VOT challenge 2014 datasets. Experimental results show that the accuracy, robustness and computational complexity of our tracker have been improved. The proposed algorithm is significantly better than several state-of-art algorithms on the OTB datasets and the VOT challenge 2014 datasets. The computational efficiency of the proposed algorithm is reduced by 75% compared to compressive tracking and nearly 10% of the DSST tracker which is the winner of VOT 2014.

Video games have become an integral part of most people’s lives in recent times. This led to an abundance of data related to video games being shared online. However, this comes with issues such as incorrect ratings, reviews or anything that is being shared. Recommendation systems are powerful tools that help users by providing them with meaningful recommendations. A straightforward approach would be to predict the scores of video games based on other information related to the game. It could be used as a means to validate user-submitted ratings as well as provide recommendations. This work provides a method to predict the G-Score, that defines how good a video game is, from its trailer (video) and summary (text). We first propose models to predict the G-Score based on the trailer alone (unimodal). Later on, we show that considering information frommultiple modalities helps the models perform better compared to using information from videos alone. We created our own dataset named VGD (Video Game Dataset) and provide it along with this work. The approach mentioned here can be generalized to other multimodal datasets such as movie trailers and summaries etc. Towards the end, we talk about the shortcomings of the work and some methods to overcome them.

The following section is included:

• Index