Large Image Collections Research Articles

Training robust T1-weighted magnetic resonance imaging liver segmentation models using ensembles of datasets with different contrast protocols and liver disease etiologies

Image segmentation of the liver is an important step in treatment planning for liver cancer. However, manual segmentation at a large scale is not practical, leading to increasing reliance on deep learning models to automatically segment the liver. This manuscript develops a generalizable deep learning model to segment the liver on T1-weighted MR images. In particular, three distinct deep learning architectures (nnUNet, PocketNet, Swin UNETR) were considered using data gathered from six geographically different institutions. A total of 819 T1-weighted MR images were gathered from both public and internal sources. Our experiments compared each architecture’s testing performance when trained both intra-institutionally and inter-institutionally. Models trained using nnUNet and its PocketNet variant achieved mean Dice-Sorensen similarity coefficients>0.9 on both intra- and inter-institutional test set data. The performance of these models suggests that nnUNet and PocketNet liver segmentation models trained on a large and diverse collection of T1-weighted MR images would on average achieve good intra-institutional segmentation performance.

PrivateGaze: Preserving User Privacy in Black-box Mobile Gaze Tracking Services

Eye gaze contains rich information about human attention and cognitive processes. This capability makes the underlying technology, known as gaze tracking, a critical enabler for many ubiquitous applications and has triggered the development of easy-to-use gaze estimation services. Indeed, by utilizing the ubiquitous cameras on tablets and smartphones, users can readily access many gaze estimation services. In using these services, users must provide their full-face images to the gaze estimator, which is often a black box. This poses significant privacy threats to the users, especially when a malicious service provider gathers a large collection of face images to classify sensitive user attributes. In this work, we present PrivateGaze, the first approach that can effectively preserve users' privacy in black-box gaze tracking services without compromising gaze estimation performance. Specifically, we proposed a novel framework to train a privacy preserver that converts full-face images into obfuscated counterparts, which are effective for gaze estimation while containing no privacy information. Evaluation on four datasets shows that the obfuscated image can protect users' private information, such as identity and gender, against unauthorized attribute classification. Meanwhile, when used directly by the black-box gaze estimator as inputs, the obfuscated images lead to comparable tracking performance to the conventional, unprotected full-face images.

Ensemble CNN Model for Computer-Aided Knee Osteoarthritis Diagnosis

Osteoarthritis (OA) is a multifaceted ailment posing challenges in its diagnosis and treatment due to the intricate nature of the disease. Particularly, Knee Osteoarthritis (KOA) significantly impacts the knee joint, manifesting through symptoms such as pain, stiffness, and limited movement. Despite its prevalence and debilitating effects, early detection of KOA remains elusive, often hindered by subjective diagnostic methods and the absence of reliable biomarkers. This research aims to address these challenges by leveraging deep learning techniques and ensemble methodologies for accurate KOA classification using knee X-ray images. This paper utilized a dataset sourced from the Osteoarthritis Initiative (OAI), comprising a large collection of knee X-ray images graded according to the Kellgren-Lawrence (KL) grading system. The proposed design methodology involves preprocessing the input X-ray images and training multiple pre-trained Convolutional Neural Network (CNN) models, including ResNet50, InceptionResNetV2, and Xception to classify KOA severity grades. Additionally, this work introduced an ensemble model by combining predictions from these base models to improve overall performance of the Computer-Aided Diagnosis (CAD) system. The obtained results demonstrate the effectiveness of the ensemble approach, outperforming individual algorithms in terms of accuracy, precision, recall, F1-score, and balanced accuracy. However, challenges persist in accurately distinguishing between adjacent KL grades, particularly grades#1 and #2, highlighting the need for further refinement. Notably, the proposed CAD model showcases superior predictive accuracy compared to various state-of-art methods, offering a promising avenue for early KOA diagnosis and personalized treatment strategies.

Advanced Deep Learning and Nonlinear Mathematical Analysis for Precision Detection and Targeted Treatment of Plant Diseases

Plant diseases are a big problem in agriculture because they lower yields around the world. Regular ways of finding and treating diseases usually require a lot of work, take a long time, and aren't always accurate. For these problems, this study shows a combined method using advanced deep learning methods like convolutional neural networks (CNNs), long short-term memory networks (LSTMs), and MobileNet, along with nonlinear mathematical analysis to improve the accuracy of finding plant diseases and making the best targeted treatment plans. Image-based diagnosis of plant diseases using deep learning methods, especially CNNs, is a quick and accurate method. They learn to spot tiny visual clues that point to different diseases by looking at large collections of images of healthy and sick plants. Addition of LSTMs and MobileNet with optimization improves the model's ability to handle time cycles and complex patterns, making it more reliable and usable for a wider range of plant types and weather conditions. The deep learning models are also fine-tuned using an optimization method, like the Adam optimizer, which ensures the best performance and faster resolution. Additionally, complex mathematical models are created to show how plant diseases spread and get worse in crop fields. Based on things like disease transfer rates, weather conditions, and plant chemistry, these models can predict how diseases will spread and help with treatment plans. This system gives dynamic, site-specific advice on how to control diseases by combining these models with real-time data from field devices and remote sensing technologies. Testing the suggested method's effectiveness in the field on many different crops shows that it works much better than traditional methods at both finding problems and treating them.

MatUp: Repurposing Image Upsamplers for SVBRDFs

AbstractWe propose MatUp, an upsampling filter for material super‐resolution. Our method takes as input a low‐resolution SVBRDF and upscales its maps so that their rendering under various lighting conditions fits upsampled renderings inferred in the radiance domain with pre‐trained RGB upsamplers. We formulate our local filter as a compact Multilayer Perceptron (MLP), which acts on a small window of the input SVBRDF and is optimized using a data‐fitting loss defined over upsampled radiance at various locations. This optimization is entirely performed at the scale of a single, independent material. Doing so, MatUp leverages the reconstruction capabilities acquired over large collections of natural images by pre‐trained RGB models and provides regularization over self‐similar structures. In particular, our light‐weight neural filter avoids retraining complex architectures from scratch or accessing any large collection of low/high resolution material pairs – which do not actually exist at the scale RGB upsamplers are trained with. As a result, MatUp provides fine and coherent details in the upscaled material maps, as shown in the extensive evaluation we provide.

EndoViT: pretraining vision transformers on a large collection of endoscopic images

PurposeAutomated endoscopy video analysis is essential for assisting surgeons during medical procedures, but it faces challenges due to complex surgical scenes and limited annotated data. Large-scale pretraining has shown great success in natural language processing and computer vision communities in recent years. These approaches reduce the need for annotated data, which is of great interest in the medical domain. In this work, we investigate endoscopy domain-specific self-supervised pretraining on large collections of data.MethodsTo this end, we first collect Endo700k, the largest publicly available corpus of endoscopic images, extracted from nine public Minimally Invasive Surgery (MIS) datasets. Endo700k comprises more than 700,000 images. Next, we introduce EndoViT, an endoscopy-pretrained Vision Transformer (ViT), and evaluate it on a diverse set of surgical downstream tasks.ResultsOur findings indicate that domain-specific pretraining with EndoViT yields notable advantages in complex downstream tasks. In the case of action triplet recognition, our approach outperforms ImageNet pretraining. In semantic segmentation, we surpass the state-of-the-art (SOTA) performance. These results demonstrate the effectiveness of our domain-specific pretraining approach in addressing the challenges of automated endoscopy video analysis.ConclusionOur study contributes to the field of medical computer vision by showcasing the benefits of domain-specific large-scale self-supervised pretraining for vision transformers. We release both our code and pretrained models to facilitate further research in this direction: https://github.com/DominikBatic/EndoViT.

OPTYMALIZACJA DRZEWA DECYZYJNEGO OPARTA NA ALGORYTMIE GENETYCZNYM DO WYKRYWANIA DEMENCJI POPRZEZ ANALIZĘ MRI

Dementia is a devastating neurological disorder that affects millions of people globally, causing progressive decline in cognitive function and daily living activities. Early and precise detection of dementia is critical for optimal dementia therapy and management however, the diagnosis of dementia is often challenging due to the complexity of the disease and the wide range of symptoms that patients may exhibit. Machine learning approaches are becoming progressively more prevalent in the realm of image processing, particularly for disease prediction. These algorithms can learn to recognize distinctive characteristics and patterns that are suggestive of specific diseases by analyzing images from multiple medical imaging modalities. This paper aims to develop and optimize a decision tree algorithm for dementia detection using the OASIS dataset, which comprises a large collection of MRI images and associated clinical data. This approach involves using a genetic algorithm to optimize the decision tree model for maximum accuracy and effectiveness. The ultimate goal of the paper is to develop an effective, non-invasive diagnostic tool for early and accurate detection of dementia. The GA-based decision tree, as proposed, exhibits strong performance compared to alternative models, boasting an impressive accuracy rate of 96.67% according to experimental results.

DISCount: Counting in Large Image Collections with Detector-Based Importance Sampling

Many applications use computer vision to detect and count objects in massive image collections. However, automated methods may fail to deliver accurate counts, especially when the task is very difficult or requires a fast response time. For example, during disaster response, aid organizations aim to quickly count damaged buildings in satellite images to plan relief missions, but pre-trained building and damage detectors often perform poorly due to domain shifts. In such cases, there is a need for human-in-the-loop approaches to accurately count with minimal human effort. We propose DISCount -- a detector-based importance sampling framework for counting in large image collections. DISCount uses an imperfect detector and human screening to estimate low-variance unbiased counts. We propose techniques for counting over multiple spatial or temporal regions using a small amount of screening and estimate confidence intervals. This enables end-users to stop screening when estimates are sufficiently accurate, which is often the goal in real-world applications. We demonstrate our method with two applications: counting birds in radar imagery to understand responses to climate change, and counting damaged buildings in satellite imagery for damage assessment in regions struck by a natural disaster. On the technical side we develop variance reduction techniques based on control variates and prove the (conditional) unbiasedness of the estimators. DISCount leads to a 9-12x reduction in the labeling costs to obtain the same error rates compared to naive screening for tasks we consider, and surpasses alternative covariate-based screening approaches.

Improving photoacoustic imaging through the skull using deep learning: a numerical study

Photoacoustic computed tomography (PACT) has recently emerged as an attractive imaging modality for functional brain imaging due to its rich optical absorption contrast, high spatial and temporal resolutions, and relatively deep penetration. However, a major hurdle in using PACT for the human brain is distortion of the signal due to the skull, which negatively affects the quality of the images. In this project, we aimed to improve transcranial PACT using a U-Net architecture that can minimize distortion from the skull. This numerical study utilized a large collection of blood vessel images obtained from an online database and a computed tomography (CT) scan of an ex vivo human skull. The synthetic photoacoustic radiofrequency data were generated using the open-source wave solver k-Wave. Comparing the images generated by deep learning with the ground truth images, we achieved an average structural similarity index of 0.874 and an average peak signal-to-noise ratio of 17.92 dB.

Enhancing skin cancer classification using a fusion of Densenet and Mobilenet models: a deep learning ensemble approach

Skin cancer is a condition that causes the formation of abnormal cells in the skin tissue. Every year, millions of individuals experience skin cancer. Skin cancer identification in its early stages is an expensive and difficult process. Currently, deep learning models have revealed promising results in automated skin cancer classification. The goal of this study is to identify benign or malignant skin lesion using the Internation Skin Imaging Collaboration (ISIC) dataset and a concatenated model that combines two powerful Convolutional Neural Network (CNN) architectures, DenseNet and MobileNet. The ISIC dataset, which consists of a large collection of dermoscopic images, provides a valuable resource for training and evaluating skin cancer classification models. By concatenating the feature maps of DenseNet and MobileNet, the proposed model capitalizes on their individual strengths. The concatenated model is trained using a combination of techniques called transfer learning and fine-tuning, which leverage the pre-trained weights of the individual models and adapt them to the skin cancer classification task. On the test dataset, our proposed model finally achieves 93.75% accuracy, which is 4.69% higher than DenseNet, 3.13% higher than MobileNet, and 3.13% higher than the combination of ResNet50 and InceptionNetV3. The execution time of our proposed ensemble method is 49.48 min, and the execution times of MobileNet and DenseNet are 178.9 and 170.5 min, which are more than our proposed ensemble approach.

PanorAMS: Automatic Annotation for Detecting Objects in Urban Context

Large collections of geo-referenced panoramic images are freely available for cities across the globe, as well as detailed maps with location and meta-data on a great variety of urban objects. They provide a potentially rich source of information on urban objects, but manual annotation for object detection is expensive, laborious and challenging. Can we utilize such multimedia sources to automatically annotate street level images as an inexpensive alternative to manual labeling? With the PanorAMS framework we introduce a method to automatically generate bounding box annotations for panoramic images based on urban context information. Following this method, we acquire large-scale, albeit noisy, annotations for an urban dataset solely from open data sources in a fast and automatic manner. The dataset covers the City of Amsterdam and includes over 14 million noisy bounding box annotations of 22 object categories present in 771,299 panoramic images. For many objects further fine-grained information is available, obtained from geospatial meta-data, such as <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">building value</i> , <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">function</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">average surface area</i> . Such information would have been difficult, if not impossible, to acquire via manual labeling based on the image alone. For detailed evaluation, we introduce an efficient crowdsourcing protocol for bounding box annotations in panoramic images, which we deploy to acquire 147,075 ground-truth object annotations for a subset of 7,348 images, the PanorAMS-clean dataset. For our PanorAMS-noisy dataset, we provide an extensive analysis of the noise and how different types of noise affect image classification and object detection performance. We make both datasets, PanorAMS-noisy and PanorAMS-clean, benchmarks and tools presented in this paper openly available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://panorams.inskegroenen.nl/</uri> .

Self-supervised contrastive learning of radio data for source detection, classification and peculiar object discovery

Abstract New advancements in radio data post-processing are underway within the Square Kilometre Array (SKA) precursor community, aiming to facilitate the extraction of scientific results from survey images through a semi-automated approach. Several of these developments leverage deep learning methodologies for diverse tasks, including source detection, object or morphology classification, and anomaly detection. Despite substantial progress, the full potential of these methods often remains untapped due to challenges associated with training large supervised models, particularly in the presence of small and class-unbalanced labelled datasets. Self-supervised learning has recently established itself as a powerful methodology to deal with some of the aforementioned challenges, by directly learning a lower-dimensional representation from large samples of unlabelled data. The resulting model and data representation can then be used for data inspection and various downstream tasks if a small subset of labelled data is available. In this work, we explored contrastive learning methods to learn suitable radio data representations by training the SimCLR model on large collections of unlabelled radio images taken from the ASKAP EMU and SARAO MeerKAT GPS surveys. The resulting models were fine-tuned over smaller labelled datasets, including annotated images from various radio surveys, and evaluated on radio source detection and classification tasks. Additionally, we employed the trained self-supervised models to extract features from radio images, which were used in an unsupervised search for objects with peculiar morphology in the ASKAP EMU pilot survey data. For all considered downstream tasks, we reported the model performance metrics and discussed the benefits brought by self-supervised pre-training, paving the way for building radio foundational models in the SKA era.

Pixel-Perfect Structure-From-Motion With Featuremetric Refinement.

Finding local features that are repeatable across multiple views is a cornerstone of sparse 3D reconstruction. The classical image matching paradigm detects keypoints per-image once and for all, which can yield poorly-localized features and propagate large errors to the final geometry. In this paper, we refine two key steps of structure-from-motion by a direct alignment of low-level image information from multiple views: we first adjust the initial keypoint locations prior to any geometric estimation, and subsequently refine points and camera poses as a post-processing. This refinement is robust to large detection noise and appearance changes, as it optimizes a featuremetric error based on dense features predicted by a neural network. This significantly improves the accuracy of camera poses and scene geometry for a wide range of keypoint detectors, challenging viewing conditions, and off-the-shelf deep features. Our system easily scales to large image collections, enabling pixel-perfect crowd-sourced localization at scale. Our code is publicly available athttps://github.com/cvg/pixel-perfect-sfm as an add-on to the popular Structure-from-Motion software COLMAP.

A deep learning outline aimed at prompt skin cancer detection utilizing gated recurrent unit networks and improved orca predation algorithm

Skin cancer is a risky ailment that can be effectively managed if detected promptly. However, timely recognition of skin cancer remains a challenge. In this study, a robotic computer-assisted tactic is proposed for the early detection of skin cancer. The motivation behind this research is to improve the accuracy and efficiency of skin cancer recognition. The proposed methodology involves a series of steps. Initially, the input images undergo preprocessing to enhance their quality and extract relevant features. These preprocessed images are then fed into a Gated Recurrent Unit (GRU) Network, a type of deep learning model known for its ability to capture sequential information. To optimize the performance of the GRU Network, we employ an enhanced variant of the Orca Predation Algorithm (OPA). This algorithm helps fine-tune the network parameters, improving its diagnostic capabilities. To validate and evaluate the effectiveness of our skin cancer diagnosis algorithm, we conducted experiments using the HAM10000 dataset, which contains a large collection of skin lesion images. We compared the results obtained from our proposed method, named GRU/IOPA, with eight existing techniques commonly used for skin cancer diagnosis. Five execution indices, namely sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy, were used to assess the performance of these methods. Our findings demonstrate that the GRU/IOPA system outperforms other existing methods in terms of sensitivity (0.95), specificity (0.97), PPV (0.95), NPV (0.96), and accuracy. These results indicate the effectiveness of our proposed method in diagnosing skin cancer compared to traditional approaches. The superior performance of GRU/IOPA highlights its potential impact on improving skin cancer diagnosis and reinforces its promise as an advanced tool in the field. In conclusion, our study presents a novel approach for timely skin cancer recognition through a robotic computer-assisted tactic. By utilizing the GRU/IOPA system, we achieve superior accuracy and efficiency in diagnosing skin cancer compared to existing techniques. This research offers significant contributions to the field of skin cancer diagnosis and opens up new avenues for future advancements in this area.

Mixed Methods Framework for Understanding Visual Frames in Social Movements

ABSTRACTAttempting to understand visual frame perspectives in social movement posts online is important to develop an account of how social movements communicate and for what purpose. This paper builds a Mixed‐Methods Matrix framework that combines computational applications with visual methodologies to discover frames of meaning making in a large image collection. Frame analysis and Critical Visual Methodology are reviewed and used in the framework to work in tangent with quantitative research methods. The quantitative methods consist of network analysis applications and network structure analysis. Visual sentiment analysis is explored using methods of computer vision. The methods framework is presented in the form of a matrix that enables researchers to identify applications for looking at social movements online through theoretical and computational approaches. The broader implication for the framework is to help researchers understand how online image collections can show meaning through perspective.

CzSL: Learning from citizen science, experts, and unlabelled data in astronomical image classification

ABSTRACT Citizen science is gaining popularity as a valuable tool for labelling large collections of astronomical images by the general public. This is often achieved at the cost of poorer quality classifications made by amateur participants, which are usually verified by employing smaller data sets labelled by professional astronomers. Despite its success, citizen science alone will not be able to handle the classification of current and upcoming surveys. To alleviate this issue, citizen science projects have been coupled with machine learning techniques in pursuit of a more robust automated classification. However, existing approaches have neglected the fact that, apart from the data labelled by amateurs, (limited) expert knowledge of the problem is also available along with vast amounts of unlabelled data that have not yet been exploited within a unified learning framework. This paper presents an innovative learning methodology for citizen science capable of taking advantage of expert- and amateur-labelled data, featuring a transfer of labels between experts and amateurs. The proposed approach first learns from unlabelled data with a convolutional auto-encoder and then exploits amateur and expert labels via the pre-training and fine-tuning of a convolutional neural network, respectively. We focus on the classification of galaxy images from the Galaxy Zoo project, from which we test binary, multiclass, and imbalanced classification scenarios. The results demonstrate that our solution is able to improve classification performance compared to a set of baseline approaches, deploying a promising methodology for learning from different confidence levels in data labelling.

Deep Learning Based Quality Prediction of Retinal Fundus Images

Abstract The accuracy of diagnosing and monitoring eye diseases using fundus imaging is strongly dependent on the quality of the images. Poor image quality can result in delays or inaccuracies in diagnosis, thus risking patient health. Image quality assessment (IQA) is crucial for regulating retinal imaging quality and assisting in successful diagnoses. In this paper, a learning-based model for evaluating the quality of fundus images is proposed. The EyeQ dataset is selected due to its ease of availability and large collection of images. The images in the dataset are classified into three labels based on their perceptual quality. Label 0 denotes good, 1 denotes fair and 2 denotes poor quality. The ResNet50 deep learning model is fine-tuned with different hyperparameters to develop two models, which are labelled as network-1 and network-2. Network-1 uses the Stochastic Gradient Descent (SGD) optimizer and network-2 utilizes the Adam optimizer. Both networks utilise the Categorical Cross Entropy (CCE) loss function, are trained for 40 epochs and has a learning rate of 1e-3. The performance of the trained models is compared using the accuracy validation curve and the loss validation curve. Network-1 demonstrates an accuracy value of 85.8% with a loss of 0.37 and network-2 achieves an accuracy of 86.2% with a loss of 0.34. The results indicate that network-2 slightly outperforms the performance of the other in terms of accuracy, and should be preferred. Moreover, it is evident that the ResNet50 model is highly adept in appropriate feature extraction and quality evaluation of retinal fundus imaging.

NeRVI: Compressive neural representation of visualization images for communicating volume visualization results

We present NeRVI, a new deep-learning approach that compresses a large collection of visualization images generated from time-varying data for communicating volume visualization results. Based on an image-based implicit neural representation, our approach represents tens of thousands of high-resolution rendering images parameterized by different parameters via a hybrid model of multilayer perceptrons and convolutional neural networks. Our model predicts images and corresponding masks, and the masks are utilized for loss computation and network training to capture fine structural details and small components. In conjunction with model quantization and weight encoding, NeRVI yields highly compact compressive neural representations while preserving the image fidelity well. We demonstrate the effectiveness of NeRVI with isosurface rendering and direct volume rendering images generated from multiple data sets and compare NeRVI with other state-of-the-art deep learning-based (InSituNet, SIREN, NeRF, and NeRV) methods. Quantitative and qualitative results show that NeRVI provides an alternative solution that augments domain scientists’ ability to manage, represent, and communicate scientific visualization output.

Counteracting Data Bias and Class Imbalance-Towards a Useful and Reliable Retinal Disease Recognition System.

Multiple studies presented satisfactory performances for the treatment of various ocular diseases. To date, there has been no study that describes a multiclass model, medically accurate, and trained on large diverse dataset. No study has addressed a class imbalance problem in one giant dataset originating from multiple large diverse eye fundus image collections. To ensure a real-life clinical environment and mitigate the problem of biased medical image data, 22 publicly available datasets were merged. To secure medical validity only Diabetic Retinopathy (DR), Age-Related Macular Degeneration (AMD) and Glaucoma (GL) were included. The state-of-the-art models ConvNext, RegNet and ResNet were utilized. In the resulting dataset, there were 86,415 normal, 3787 GL, 632 AMD and 34,379 DR fundus images. ConvNextTiny achieved the best results in terms of recognizing most of the examined eye diseases with the most metrics. The overall accuracy was 80.46 ± 1.48. Specific accuracy values were: 80.01 ± 1.10 for normal eye fundus, 97.20 ± 0.66 for GL, 98.14 ± 0.31 for AMD, 80.66 ± 1.27 for DR. A suitable screening model for the most prevalent retinal diseases in ageing societies was designed. The model was developed on a diverse, combined large dataset which made the obtained results less biased and more generalizable.

Lexicon-based probabilistic indexing of handwritten text images

Keyword Spotting (KWS) is here considered as a basic technology for Probabilistic Indexing (PrIx) of large collections of handwritten text images to allow fast textual access to the contents of these collections. Under this perspective, a probabilistic framework for lexicon-based KWS in text images is presented. The presentation aims at providing formal insights which help understanding classical statements of KWS (from which PrIx borrows fundamental concepts), as well as the relative challenges entailed by these statements. The development of the proposed framework makes it clear that word recognition or classification implicitly or explicitly underlies any formulation of KWS. Moreover, it suggests that the same statistical models and training methods successfully used for handwriting text recognition can advantageously be used also for PrIx, even though PrIx does not generally require or rely on any kind of previously produced image transcripts. Experiments carried out using these approaches support the consistency and the general interest of the proposed framework. Results on three datasets traditionally used for KWS benchmarking are significantly better than those previously published for these datasets. In addition, good results are also reported on two new, larger handwritten text image datasets (Bentham and Plantas), showing the great potential of the methods proposed in this paper for indexing and textual search in large collections of untranscribed handwritten documents. Specifically, we achieved the following Average Precision values: IAMDB: 0.89, George Washington: 0.91, Parzival: 0.95, Bentham: 0.91 and Plantas: 0.92.