Advances in Pattern Recognition and Artificial Intelligence

The following sections are included:

• Preface
• Contents
• List of Contributors

Breast cancer is the most common cancer in women worldwide. Computer Aided Detection (CADe) has attracted increasing research interest in recent years. Data exploration and lesion detection in medical images are tedious but can be accelerated using computational intelligence. One approach is to adapt and configure recent deep learning-based object detectors from computer vision to detect abnormalities in medical images. This chapter starts with three state-of-the-art object detectors, namely Faster R-CNN, YOLO v2 and Grad-CAM, to determine the location of lesions in Magnetic Resonance Images (MRI) of breast. Each individual detector is first tuned through adjusting network hyper-parameters and backbone architectures, in order to maximize Average Precision (AP). Two different integration approaches, namely cascaded and parallel integrations, are then proposed and implemented to improve the AP. The cascaded integration builds on a coarse-to-fine strategy. It uses Grad-CAM to compute coarse locations of bounding boxes for lesions and then applies YOLO v2 or Faster R-CNN as a fine detector. In the parallel integration approach, the detection result is a combination of results from the three detectors, with the aim of reducing missed detections. The integrated deep learning models exhibit enhanced performance on the detection of breast lesions in MRIs.

Peripheral Blood Smear (PBS) analysis is a routine test carried out in specialized medical laboratories by specialists to assess some aspects of health status that are measured and assessed through blood. PBS analysis is prone to human errors and the usage of computer-based analysis can greatly enhance this process in terms of accuracy and cost. Despite the challenges, Deep Learning neural networks have shown impressive performance in this context. In this study the recent contributions are summarized along with the main challenges and future directions in this context.

Although it has received increasing researchers’ attention in recent years, palm-vein recognition still faces various challenges in practical applications, one of which is the lack of robustness against image quality degradation, resulting in reduction of the verification accuracy. To address this problem, this chapter proposes an end-to-end generative adversarial network to automatically extract the vein pattern network, thus without resorting to any hand-crafted segmentation techniques of grayscale images into vein pixels and background. Firstly, we label the palm-vein pixels based on a combination of handcrafted segmentation methods and reconstruct a training set accordingly. Secondly, an end-to-end vein segmentation model is proposed based on a generative adversarial network. After training, this model outputs an image map where each value is the probability that the corresponding pixel belongs to a vein pattern. The resulting map is then subject to binarization by thresholding and stored in a binary image, used subsequently for verification matching. The experimental results on the public CASIA and PolyU palm-vein datasets demonstrate the effectiveness of our proposed method.

As deep learning methods are being used increasingly in the field of retinal image analysis, the challenges of insufficient and imbalanced dataset begin to affect the performance of the retinal image analysis models. In this chapter, we introduce a training method which uses data augmentation to overcome the insufficiency and imbalance of retinal image dataset, particularly the retinal image for the classification of diabetic retinopathy (DR). By using the data augmentation method, we are able to obtain a balanced augmented retinal image dataset from the original imbalanced retinal image dataset to expand the size of the original retinal dataset about four times, leading to better classification accuracy.

This article discusses handwriting as a means of personality assessment. Through a discussion of the various internal and external factors, we see how the elements of spatial arrangement, writing form, and writing movement develop handwriting patterns. These are shown to combine in ways that demonstrate a wide range of personality traits in handwriting. The question is addressed whether cursive writing is still important in a digital age.

A writing system contains several patterns/graphical symbols, which represent linguistic constructs. Such patterns/symbols are orthographically related to the phonology of a spoken language. A writing system may contain a few thousands of such symbols (e.g., about 10,000 frequently-used symbols for the Chinese writing system, about 1000 for Devanagari, etc.). Therefore, a section of people face difficulties in learning and writing all the character structures of a writing system. Besides, due to the dominance of electronic typing in this digital age, individuals often experience increasing difficulty in writing the characters by hand, even though they can read those. In this chapter, we analyze the handwriting of an individual and attempt to understand which character shapes are challenging for that person to write. We propose here a deep reinforcement learning-based model to predict the difficulty measure/score of a character written by an individual. For experimentation, we have chosen Bengali, a script of the Indian subcontinent, which contains about 1000 complex character shapes. We have obtained some promising results from our experiments.

Offline gender detection from Arabic handwritten documents is a very challenging task because of the high similarity between an individual’s writings and the complexity of the Arabic language as well. In this chapter, we propose a new way to detect the writer gender from scanned handwritten documents that mainly based on the concept of transfer-learning. We used a pre-trained knowledge from two convolution neural networks (CNN): GoogleNet, and ResNet, then we applied it on our dataset. We use this two CNN architectures as fixed feature extractors. For the analysis and the classification stage, we used a support vector machine (SVM). The performance of the two CNN architectures concerning accuracy is 80.05% for GoogleNet, 83.32% for ResNet.

The Wartegg test is a useful semi-projective drawing test for counseling professionals. The advancement of research in artificial intelligence opens up new perspectives in the analysis and interpretation of the test. An alliance with the research group led by the Italian psychologist Alessandro Crisi could be very profitable.

Credit scoring prediction technique is to assist the lender in making decisions. Accurately predicting user’s credit scoring is a challenging issue. In this chapter, a new credit scoring prediction method is proposed, in which the users and their relationships are modeled as a weighted undirected graph, and then the credit scoring prediction problem is reduced to a graph signal reconstruction problem. The proposed method utilizes both the information of the unlabeled data and the location information of the samples in the feature space, and thus achieves an excellent predictive performance. Experimental results on the open UCI German credit dataset are compared with those of seven classical models that the prediction performance of the proposed model is significantly better than that of the reference models. The Friedman test indicates that the experimental results have statistical significance.

This document discusses about the notion of features in image analysis. With feature we designate here some information issued from the raw image data, related to a visual pattern, that is supposed to have some interest for a particular task from the pattern recognition pipeline. After discussing about the what for, whom for and how, we present some examples from the literature highlighting the high diversity of features and some of their properties, which can be worth considering both by the computer scientist and the expert when designing an image analysis system. We finally provide some potential take-away messages to the reader, arguing that an optimal choice of features is rarely handleable since such a choice is more often a compromise between computation and application needs.

In this chapter, information about a large and diverse database of license plates from countries around the world is presented. CENPARMI’s growing database contains images of isolated plates, including a small percentage of vanity plates, as well as landscapes, where the vehicle is included in the image. Many images contain multiple license plates, have complex scenery, have motion blur, and contain high light and shadow contrast. Photos were taken during different seasons, many are occluded by foreground objects, and some were taken through mist, fog, snow, and/or glass. In many cases, the license plate, camera, or both were in motion. In order to make training more robust, different cameras, lenses, focal lengths, shutter speeds, and ISOs (sensor sensitivity) were used. A summary of proposed guidelines used for license plate design is outlined, details about the database of license plates are presented, followed by information about ongoing work and efforts for labelling/ground truth.

With the rapid expansion of digital newspapers, readers have an overwhelming amount of news available daily. However, it is difficult to keep track of the news that is only of interest to the reader or filtering those that belong to a specific topic or crime, e.g. assault or kidnapping. The main challenge is handling free texts from Spanish criminal news; however, deep learning has proposed models of representation and manipulation of natural language that help in this problem. Therefore, this chapter proposes the use of deep neural models to classify criminal news published by Mexican digital newspapers. The whole approach includes a word embedding model for representing the documents and several layers in neural networks to classify news text. We experiment with three deep neural models: Bidirectional LSTM, Attention Bidirectional LSTM and BERT-based model. According to the experimental results, our system shows promising results.

Increasingly, politicians and political parties are engaging their electors using social media. In the US Federal Election of 2016, candidates from both parties made heavy use of Social Media, particularly Twitter. It is then reasonable to attempt to find a correlation between popularity on Twitter, and eventual popular vote in the election. In this study, we will focus on using the subscriber ‘location’ field in the profile of each candidate to estimate support in each state.

In this Study, we will train a Deep Convolutional Neural Network (CNN) to classify place names by state. Then we will apply the model to the Twitter Subscriber ‘location’ field of Twitter subscribers collected from each of the two candidates, Hillary Clinton (D), and Donald Trump (R). Finally, we will compare predicted popular votes in each state, to the actual results from the 2016 Presidential Election.

The hypothesis is that a city name has a strong correlation to the people who founded it and then incorporated it. Further, it is hypothesized that the original settlers were mostly homogeneous, relative to the country of origin and shared a common language, thus resulting in place names using the language of their origin. The results from our experiments are very promising.

The following section is included:

• Index