This paper explores an effective method for retrieving additional images of a specific individual from large, unannotated photo collections using a reference query image. This task, known as face image retrieval (FIR), focuses on maximising recall by increasing the proportion of correct matches and improving precision by retrieving a greater number of relevant images. However, precision and recall inherently contradict each other, meaning that an increase in one often leads to a decrease in the other. To enhance the retrieval process, two key improvements are introduced. First, the existing consensus ranking method is strengthened to perform more reliably and faster. Second, it is discovered that mirroring the query image and averaging the corresponding face feature vectors leads to an overall improvement in both precision and recall. The proposed method is evaluated using both a historical photo dataset derived from criminal identification photos and a publicly available dataset for face recognition. Additionally, the efficiency of the CVLFace pipeline is examined, further assessing the overall effectiveness of the approach. These refinements contribute to a more robust approach for searching unannotated face photo databases, ensuring more effective retrieval performance.

One of the main problems for face recognition when comparing photos of various ages is the impact of age progression on facial features. The face undergoes many changes as a person grows older, including geometrical changes and changes in facial hair, etc. Even though biometric markers such as computed face feature vectors should preferably be invariant to such factors, face recognition generally becomes less reliable as the age span grows larger. Therefore, this study was conducted with the aim of exploring the efficiency of such feature vectors in recognising individuals despite variations in age, and how to measure face recognition performance and behaviour in the data. It is shown that they are indeed discriminative enough to achieve age-invariant face recognition without synthesising age images through generative processes or training on specialised age related features.

One of the major difficulties in face recognition while comparing photographs of individuals of different ages is the influence of age progression on their facial features. As a person ages, the face undergoes many changes, such as geometrical changes, changes in facial hair, and the presence of glasses, among others. Although biometric markers like computed face feature vectors should ideally remain unchanged by such factors, face recognition becomes less reliable as the age range increases. Therefore, this investigation was carried out to examine how the use of Embedded Prototype Subspace Classifiers could improve face recognition accuracy when dealing with age-related variations using face feature vectors only.

Efficient face image retrieval, i.e. searching for existing photographs of a person in unlabelled photo collections using a query photo, is evaluated for a novel method to find the top n results for Consensus Ranking. The approach aims to maximise precision and recall by using the retrieved photos, all ranked on similarity. The proposed method aims to retrieve all photos of the queried person while excluding images of other individuals. To achieve this, the method uses the top n results as temporary queries, recalculates similarities, and combines the obtained ranked lists to produce a better overall ranking. The method includes a novel and reliable procedure for selecting n, which is evaluated on two datasets, and considers the impact of age variation in the datasets.

In recent academic literature Sex and Gender have both become synonyms, even though distinct definitions do exist. This give rise to the question, which of those two are actually face image classifiers identifying? It will be argued and explained why CNN based classifiers will generally identify gender, while feeding face recognition feature vectors into a neural network, will tend to verify sex rather than gender. It is shown for the first time how state of the art Sex Classification can be performed using Embedded Prototype Subspace Classifiers (EPSC) and also how the projection depth can be learned efficiently. The automatic Gender classification, which is produced by the emph{InsightFace} project, is used as a baseline and compared to the results given by the EPSC, which takes the feature vectors produced by emph{InsightFace} as input. It turns out that the depth of projection needed is much larger for these face feature vectors than for an example classifying on MNIST or similar. Therefore, one important contribution is a simple method to determine the optimal depth for any kind of data. Furthermore, it is shown how the weights in the final layer can be set in order to make the choice of depth stable and independent of the kind of learning data. The resulting EPSC is extremely light weight and yet very accurate, reaching over $98\%$ accuracy for several datasets.

Transcribing struck-through, handwritten words, for example for the purpose of genetic criticism, can pose a challenge to both humans and machines, due to the obstructive properties of the superimposed strokes. This paper investigates the use of paired image to image translation approaches to remove strikethrough strokes from handwritten words. Four different neural network architectures are examined, ranging from a few simple convolutional layers to deeper ones, employing Dense blocks. Experimental results, obtained from one synthetic and one genuine paired strikethrough dataset, confirm that the proposed paired models outperform the CycleGAN-based state of the art, while using less than a sixth of the trainable parameters.

The purpose of this paper is to in detail describe and analyse a Fourier based handcrafted descriptor for word recognition. Especially, it is discussed how the Variability in the results can be analysed and visualised. This efficiency of the descriptor is evaluated for the use with embedded prototype subspace classifiers for handwritten word recognition. Nonetheless, it can be used with any classifier for any purpose. An hierarchical composition of discrete semicircles in the Fourier-space is proposed and it will will be show how this compares to Gabor filters, which can be used to extract edges in an image. In comparison to Histogram of Oriented Gradients, the proposed feature descriptor performs better in this scenario. Compression using PCA turns out to be able to increase both the F1-score as well as decreasing the Variability.

The general focus of the PhD presented in this paper lies on document image analysis and handwritten text recognition for historical documents. It was originally aimed at medieval manuscripts but has recently shifted to also include the processing of 20th century stenographic records. Besides general method development, a portion of the PhD studies has been dedicated to the development of visualisation and transcription tools.

This paper presents an approach towards word recognition based on embedded prototype subspace classification.The purpose of this paper is three-fold. Firstly, a new dataset for word recognition is presented, which is extractedfrom the Esposalles database consisting of the Barcelona cathedral marriage records. Secondly, different clusteringtechniques are evaluated for Embedded Prototype Subspace Classifiers. The dataset, containing 30 different classesof words is heavily imbalanced, and some word classes are very similar, which renders the classification task ratherchallenging. For ease of use, no stratified sampling is done in advance, and the impact of different data splits isevaluated for different clustering techniques. It will be demonstrated that the original clustering technique based onscaling the bandwidth has to be adjusted for this new dataset. Thirdly, an algorithm is therefore proposed that findskclusters, striving to obtain a certain amount of feature points in each cluster, rather than finding some clustersbased on scaling the Silverman’s rule of thumb. Furthermore, Self Organising Maps are also evaluated as both aclustering and embedding technique.

Obtaining the original, clean forms of struck-through handwritten words can be of interest to literary scholars, focusing on tasks such as genetic criticism. In addition to this, replacing struck-through words can also have a positive impact on text recognition tasks. This work presents a novel unsupervised approach for strikethrough removal from handwritten words, employing cycle-consistent generative adversarial networks (CycleGANs). The removal performance is improved upon by extending the network with an attribute-guided approach. Furthermore, two new datasets, a synthetic multi-writer set, based on the IAM database, and a genuine single-writer dataset, are introduced for the training and evaluation of the models. The experimental results demonstrate the efficacy of the proposed method, where the examined attribute-guided models achieve F_1 scores above 0.8 on the synthetic test set, improving upon the performance of the regular CycleGAN. Despite being trained exclusively on the synthetic dataset, the examined models even produce convincing cleaned images for genuine struck-through words.

Astrid Lindgren, Swedish author of children’s books, is known for having both composed and edited her literary work in the Melin system of shorthand (a Swedish shorthand system based on Gabelsberger). Her original drafts and manuscripts are preserved in 670 stenographed notepads kept at the National Library of Sweden and The Swedish Institute of Children’s Books. For long these notepads have been considered undecipherable and are until recently untouched by research. This paper introduces handwritten text recognition (HTR) and document image analysis (DIA) approaches to address the challenges inherent in Lindgren’s original drafts and manuscripts. It broadly covers aspects such as preprocessing and extraction of words, alignment of transcriptions and the fast transcription of large amounts of words. This is the first work to apply HTR and DIA to Gabelsberger-based shorthand material. In particular, it presents early-stage results which demonstrate that these stenographed manuscripts can indeed be transcribed, both manually by experts and by employing computerised approaches.

Deep learning approaches suffer from the so called interpretability problem and can therefore be very hard to visualise.Embedded Prototype Subspace Classifiers is one attempt in the field of explainable AI, which is both fast and efficient since it does not require repeated learning epochs and has no hidden layers.In this paper we investigate how ensembles and cascades of ensembles perform on some popular datasets.The focus is on handwritten data such as digits, letters and signs.It is shown how cascading can be efficiently implemented in order to both increase accuracy as well as speed up the classification.

Word spotting use a query word image to find any instances of that word among document images. The obtained list of words is ranked according to similarity to the query word. Ideally, any false pos ...

Logo and Seal serves the purpose of authenticating and referring to the source of a document. This strategy was also prevalent in the medieval period. Different algorithm exists for detection of logo and seal in document images. A close look into the present state-of-the-art methods reveals that those methods were focused toward detection of logo and seal in contemporary document images. However, such methods are likely to underperform while dealing with historical documents. This is due to the fact that historical documents are attributed with additional challenges like extra noise, bleed-through effect, blurred foreground elements and low contrast. The proposed method frames the problem of the logo and seals detection in an object detection framework. Using a deep-learning technique it counters earlier mentioned problems and evades the need for any pre-processing stage like layout analysis and/or binarization in the system pipeline. The experiments were conducted on historical images from 12th to the 16th century and the results obtained were very encouraging for detecting logo in historical document images. To the best of our knowledge, this is the first attempt on logo detection in historical document images using an object-detection based approach.

Ability to learn from a single instance is something unique to the human species and One-shot learning algorithms try to mimic this special capability. On the other hand, despite the fantastic performance of Deep Learning-based methods on various image classification problems, performance often depends having on a huge number of annotated training samples per class. This fact is certainly a hindrance in deploying deep neural network-based systems in many real-life applications like face recognition. Furthermore, an addition of a new class to the system will require the need to re-train the whole system from scratch. Nevertheless, the prowess of deep learned features could also not be ignored. This research aims to combine the best of deep learned features with a traditional One-Shot learning framework. Results obtained on 2 publicly available datasets are very encouraging achieving over 90% accuracy on 5-way One-Shot tasks, and 84% on 50-way One-Shot problems.

A framework for interactive visualization of script characteristics, as present in the form of handwritten letters, is proposed in this work. The basic idea behind this investigation is to lay the foundations for creating a comprehensive atlas over letter forms extracted from a large collection of handwritten documents, with minimal human guidance. The visualization of the results is based on the atlas metaphor and uses the t-SNE visualization method for creating island-like clusters that can be investigated using the proposed visualization framework. By changing a scale parameter one can investigate the dataset on different levels, i.e different sizes of the clusters.

Historical handwritten text recognition is an interesting yet challenging problem. In recent times, deep learning based methods have achieved significant performance in handwritten text recognition. However, handwriting recognition using deep learning needs training data, and often, text must be previously segmented into lines (or even words). These limitations constrain the application of HTR techniques in document collections, because training data or segmented words are not always available. Therefore, this paper proposes a training-free and segmentation-free word spotting approach that can be applied in unconstrained scenarios. The proposed word spotting framework is based on document query word expansion and relaxed feature matching algorithm, which can easily be parallelised. Since handwritten words posses distinct shape and characteristics, this work uses a combination of different keypoint detectors and Fourier-based descriptors to obtain a sufficient degree of relaxed matching. The effectiveness of the proposed method is empirically evaluated on well-known benchmark datasets using standard evaluation measures. The use of informative features along with query expansion significantly contributed in efficient performance of the proposed method.

Handwritten text recognition is a daunting task, due to complex characteristics of handwritten letters. Deep learning based methods have achieved significant advances in recognizing challenging handwritten texts because of its ability to learn and accurately classify intricate patterns. However, there are some limitations of deep learning, such as lack of well-defined mathematical model, black-box learning mechanism, etc., which pose challenges. This paper aims at going beyond the black-box learning and proposes a novel learning framework called as Embedded Prototype Subspace Classification, that is based on the well-known subspace method, to recognise handwritten letters in a fast and efficient manner. The effectiveness of the proposed framework is empirically evaluated on popular datasets using standard evaluation measures.