Multiple Training Images Research Articles

Label Enhancement-Based Multiscale Transformer for Palm-Vein Recognition

Vein biometrics is a high security and privacy preserving identification technology that has received increasing attentions. Although deep neural networks (DNNs), such as convolutional neural network (CNN), have been investigated for vein recognition and achieved a significant improvement in accuracy, they still fail to model long-range pixel dependencies in an image. Moreover, their performance is limited because the one-hot label vector employed for training may ignore the relevance among labels. To address these problems, we propose LE-MSVT, a Label Enhancement based Multi-Scale Vein Transformer for palm-vein recognition in this paper. First, we propose a multi-scale vein transformer (MSVT) to learn robust and multi-scale features, which consists of a convolutional block that captures the local information and a self-attention block that extracts scale dependencies among images with different scales. Second, to capture the relevance among labels, we explore a graph convolutional network based label enhancement (GCNLE) approach to recover the realistic label distribution for vein classification improvement. GCNLE exploits a multi-layer perception to learn an effective label correlation matrix for extracting the relation information between an input image and multiple training images from different classes. The label distribution vector is generated and then combined with the one-hot label to compute a realistic label distribution of the input image. Finally, we apply GCNLE to MSVT to obtain LE-MSVT, which is trained in an end-to-end way to further improve the feature representation capacity of MSVT classifier. We conduct extensive experiments in terms of MSVT performance and LE-MSVT improvements on three public palm-vein databases, and the experimental results show that the resulting MSVT outperforms other vein identification approaches and achieves the best performance among existing approaches, and GCNLE can greatly improve the performance of MSVT among other deep learning based classifiers.

Extraction of Human Motion Information from Digital Video Based on 3D Poisson Equation

Based on the 3D Poisson equation, this paper extracts the features of the digital video human body action sequence. By solving the Poisson equation on the silhouette sequence, the time and space features, time and space structure features, shape features, and orientation features can be obtained. First, we use the silhouette structure features in three-dimensional space-time and the orientation features of the silhouette in three-dimensional space-time to represent the local features of the silhouette sequence and use the 3D Zernike moment feature to represent the overall features of the silhouette sequence. Secondly, we combine the Bayesian classifier and AdaBoost classifier to learn and classify the features of human action sequences, conduct experiments on the Weizmann video database, and conduct multiple experiments using the method of classifying samples and selecting partial combinations for training. Then, using the recognition algorithm of motion capture, after the above process, the three-dimensional model is obtained and matched with the model in the three-dimensional model database, the sequence with the smallest distance is calculated, and the corresponding skeleton is outputted as the results of action capture. During the experiment, the human motion tracking method based on the university matching kernel (EMK) image kernel descriptor was used; that is, the scale invariant operator was used to count the characteristics of multiple training images, and finally, the high-dimensional feature space was mapped into the low-dimensional to obtain the feature space approximating the Gaussian kernel. Based on the above analysis, the main user has prior knowledge of the network environment. The experimental results show that the method in this paper can effectively extract the characteristics of human body movements and has a good classification effect for bending, one-foot jumping, vertical jumping, waving, and other movements. Due to the linear separability of the data in the kernel space, fast linear interpolation regression is performed on the features in the feature space, which significantly improves the robustness and accuracy of the estimation of the human motion pose in the image sequence.

Learning spatial-temporally regularized complementary kernelized correlation filters for visual tracking

Despite excellent performance shown by spatially regularized discriminative correlation filters (SRDCF) for visual tracking, some issues remain open that hinder further boosting their performance: first, SRDCF utilizes multiple training images to formulate its model, which makes it unable to exploit the circulant structure of the training samples in learning, leading to high computational burden; second, SRDCF is unable to efficiently exploit the powerfully discriminative nonlinear kernels, further negatively affecting its performance. In this paper, we present a novel spatial-temporally regularized complementary kernelized CFs (STRCKCF) based tracking approach. First, by introducing spatial-temporal regularization to the filter learning, the STRCKCF formulates its model with only one training image, which can not only facilitate exploiting the circulant structure in learning, but also reasonably approximate the SRDCF with multiple training images. Furthermore, by incorporating two types of kernels whose matrices are circulant, the STRCKCF is able to fully take advantage of the complementary traits of the color and HOG features to learn a robust target representation efficiently. Besides, our STRCKCF can be efficiently optimized via the alternating direction method of multipliers (ADMM). Extensive evaluations on OTB100 and VOT2016 visual tracking benchmarks demonstrate that the proposed method achieves favorable performance against state-of-the-art trackers with a speed of 40 fps on a single CPU. Compared with SRDCF, STRCKCF provides a 8 × speedup and achieves a gain of 5.5% AUC score on OTB100 and 8.4% EAO score on VOT2016.

Uncertainty Quantification in Reservoir Prediction: Part 2\u2014Handling Uncertainty in the Geological Scenario

Models used for reservoir prediction are subject to various types of uncertainty, and interpretational uncertainty is one of the most difficult to quantify due to the subjective nature of creating different scenarios of the geology and due to the difficultly of propagating these scenarios into uncertainty quantification workflows. Non-uniqueness in geological interpretation often leads to different ways to define the model. Uncertainty in the model definition is related to the equations that are used to describe the modelled reality. Therefore, it is quite challenging to quantify uncertainty between different model definitions, because they may include completely different model parameters. This paper is a continuation of work to capture geological uncertainties in history matching and presents a workflow to handle uncertainty in the geological scenario (i.e. the conceptual geological model) to quantify its impact on the reservoir forecasting and uncertainty quantification. The workflow is based on inferring uncertainty from multiple calibrated models, which are solutions of an inverse problem, using adaptive stochastic sampling and Bayesian inference. The inverse problem is solved by sampling a combined space of geological model parameters and a space of reservoir model descriptions, which represents uncertainty across different modelling concepts based on multiple geological interpretations. The workflow includes building a metric space for reservoir model descriptions using multi-dimensional scaling and classifying the metric space with support vector machines. The proposed workflow is applied to a synthetic reservoir model example to history match it to the known truth case reservoir response. The reservoir model was designed using a multi-point statistics algorithm with multiple training images as alternative geological interpretations. A comparison was made between predictions based on multiple reservoir descriptions and those of a single one, revealing improved performance in uncertainty quantification when using multiple training images.

An Improved Face Recognition based on Scale Invariant Feature Transform (SIFT): Training for Integrating Multiple Images and Matching by Key Point’s Descriptor-Geometry

Objectives: We proposed for reduction of the computational complexity and improvement of the recognition precision in the face recognition system using Scale Invariant Feature Transform (SIFT) local feature approaches. Methods/Statistical Analysis: The first one is a novel training procedure for the integration of multiple training images. This training procedure performs to remove redundant local features and blend different local features from face image. The second one is a proposed matching scheme not only considering the similarity of key point’s descriptor but also the geometry property through one-to-one matching between query and references images. Findings: This research finds the optimal settings of parameter for the proposed face recognition system based on SIFT. First, we have analyzed the change of recognition rate according to the resolution of face image in the proposed system. Then, to effect of the reduced number of key points per subject on the recognition rate and the resolution of face image were analyzed with the multiple templates per subject. As a result, we observed that the proposed template training procedure using Lowe’s key points detection method with 50×61 resolution of face images achieves higher recognition rate than the holistic approaches. The usage of Geng’s key point detection method in the proposed system obtains higher recognition rate than the usage of Lowe’s method. From the experimental result with ORL databases, the proposed face recognition system gives 99.5% of rate, which shows the higher performance than the previous ones. In addition, the proposed integration method of multiple training images reduces the number of key points by average 49.84% than the method of using multiple templates. Improvements/Applications: The experimental result of the proposed system in two well-known face databases shows that the computational quantities are reduced effectively compared to other SIFT based methods, and it gives better performance on face recognition accuracy.

Largest Matching Areas for Illumination and Occlusion Robust Face Recognition.

In this paper, we introduce a novel approach to face recognition which simultaneously tackles three combined challenges: (1) uneven illumination; (2) partial occlusion; and (3) limited training data. The new approach performs lighting normalization, occlusion de-emphasis and finally face recognition, based on finding the largest matching area (LMA) at each point on the face, as opposed to traditional fixed-size local areabased approaches. Robustness is achieved with novel approaches for feature extraction, LMA-based face image comparison and unseen data modeling. On the extended YaleB and AR face databases for face identification, our method using only a single training image per person, outperforms other methods using a single training image, and matches or exceeds methods which require multiple training images. On the labeled faces in the wild face verification database, our method outperforms comparable unsupervised methods. We also show that the new method performs competitively even when the training images are corrupted.

Refined codebook for grayscale image coding based on vector quantization

Vector quantization (VQ) is a commonly used technique for image compression. Typically, the common codebooks (CCBs) that are designed by using multiple training images are used in VQ. The CCBs are stored in the public websites such that their storage cost can be omitted. In addition to the CCBs, the private codebooks (PCBs) that are designed by using the image to be compressed can be used in VQ. However, calculating the bit rates (BRs) of VQ includes the storage cost of the PCBs. It is observed that some codewords in the CCB are not used in VQ. The codebook refinement process is designed to generate the refined codebook (RCB) based on the CCB of each image. To cut down the BRs, the lossless index coding process and the two-stage lossless coding process are employed to encode the index table and the RCB, respectively. Experimental results reveal that the proposed scheme (PS) achieves better image qualities than VQ with the CCBs. In addition, the PS requires less BRs than VQ with the PCBs.

Spatial modelling of geological domains with multiple training images: Application to the Red Dog mine, Alaska, United States

Spatial modelling of geological domains with multiple training images: Application to the Red Dog mine, Alaska, United States

Personal identification based on multiple keypoint sets of dorsal hand vein images

This paper presents a biometric identification system based on near-infrared imaging of dorsal hand veins and matching of the keypoints that are extracted from the dorsal hand vein images by the scale-invariant feature transform. The whole system is covered in detail, which includes the imaging device used, image processing methods proposed for geometric correction, region-of-interest extraction, image enhancement and vein pattern segmentation, as well as image classification by extraction and matching of keypoints. In addition to several constraints introduced to minimise incorrectly matched keypoints, a particular focus is placed on the use of multiple training images of each hand class to improve the recognition performance for a large database with more than 200 hand classes. By organising multiple keypoint sets extracted from multiple training images of each hand class into three sets, namely, the union, the intersection and the exclusion, based on their inter-class and intra-class relationships, this study shows the contribution made by each set to the recognition performance and demonstrates the feasibility of achieving 100% correct recognition by combining the three sets, based on the experiments conducted using more than 2000 dorsal hand vein images.

A Multiple Training Image Approach for Spatial Modeling of Geologic Domains

Characterization of complex geological features and patterns remains one of the most challenging tasks in geostatistics. Multiple point statistics (MPS) simulation offers an alternative to accomplish this aim by going beyond classical two-point statistics. Reproduction of features in the final realizations is achieved by borrowing high-order spatial statistics from a training image. Most MPS algorithms use one training image at a time chosen by the geomodeler. This paper proposes the use of multiple training images simultaneously for spatial modeling through a scheme of data integration for conditional probabilities known as a linear opinion pool. The training images (TIs) are based on the available information and not on conceptual geological models; one image comes from modeling the categories by a deterministic approach and another comes from the application of conventional sequential indicator simulation. The first is too continuous and the second too random. The mixing of TIs requires weights for each of them. A methodology for calibrating the weights based on the available drillholes is proposed. A measure of multipoint entropy along the drillholes is matched by the combination of the two TIs. The proposed methodology reproduces geologic features from both TIs with the correct amount of continuity and variability. There is no need for a conceptual training image from another modeling technique; the data-driven TIs permit a robust inference of spatial structure from reasonably spaced drillhole data.

Adaptive Conditioning of Multiple-Point Statistical Facies Simulation to Flow Data with Probability Maps

Multiple-point statistics (MPS) provides a flexible grid-based approach for simulating complex geologic patterns that contain high-order statistical information represented by a conceptual prior geologic model known as a training image (TI). While MPS is quite powerful for describing complex geologic facies connectivity, conditioning the simulation results on flow measurements that have a nonlinear and complex relation with the facies distribution is quite challenging. Here, an adaptive flow-conditioning method is proposed that uses a flow-data feedback mechanism to simulate facies models from a prior TI. The adaptive conditioning is implemented as a stochastic optimization algorithm that involves an initial exploration stage to find the promising regions of the search space, followed by a more focused search of the identified regions in the second stage. To guide the search strategy, a facies probability map that summarizes the common features of the accepted models in previous iterations is constructed to provide conditioning information about facies occurrence in each grid block. The constructed facies probability map is then incorporated as soft data into the single normal equation simulation (snesim) algorithm to generate a new candidate solution for the next iteration. As the optimization iterations progress, the initial facies probability map is gradually updated using the most recently accepted iterate. This conditioning process can be interpreted as a stochastic optimization algorithm with memory where the new models are proposed based on the history of the successful past iterations. The application of this adaptive conditioning approach is extended to the case where multiple training images are proposed as alternative geologic scenarios. The advantages and limitations of the proposed adaptive conditioning scheme are discussed and numerical experiments from fluvial channel formations are used to compare its performance with non-adaptive conditioning techniques.

Illumination-insensitive texture discrimination based on illumination compensation and enhancement

As the appearance of a 3D surface texture is strongly dependent on the illumination direction, 3D surface-texture classification methods need to employ multiple training images captured under a variety of illumination conditions for each class. Texture images under different illumination conditions and directions still present a challenge for texture-image retrieval and classification. This paper proposes an efficient method for illumination-insensitive texture discrimination based on illumination compensation and enhancement. Features extracted from an illumination-compensated or -enhanced texture are insensitive to illumination variation; this can improve the performance for texture classification. The proposed scheme learns the average illumination-effect matrix for image representation under changing illumination so as to compensate or enhance images and to eliminate the effect of different and uneven illuminations while retaining the intrinsic properties of the surfaces. The advantage of our method is that the assumption of a single-point light source is not required, so it circumvents and overcomes the limitations of the Lambertian model and is also suitable for outdoor settings. We use a wide range of textures in the PhoTex database in our experiments to evaluate the performance of the proposed method. Experimental results demonstrate the effectiveness of our proposed methods.

Robust intelligent PCA‐based face recognition framework using GNP‐fuzzy data mining

AbstractTraditional principal component analysis (PCA) based face recognition algorithms have a low recognition accuracy due to the influence of noise and illumination changes. This paper proposes a robust, intelligent PCA‐based face recognition framework in the complicated illumination database when using multiple training images per person (MTIP‐CID). There are mainly two improvements in the proposed method. One is that a face‐recognition‐oriented genetic‐based clustering algorithm is introduced to reduce the influence of a large number of classes on the classification accuracy in the MTIP‐CID. The other is that a classifier based on fuzzy class association rules (FCARs) is applied to mine the inherent relationships between eigenfaces and to improve the robustness of PCA‐based face recognition in noisy environments. Experimental results on the extended Yale‐B database demonstrate that the proposed framework performs better and is more robust against noise compared with other traditional face recognition algorithms, i.e. linear discriminant analysis (LDA) and local binary patterns (LBPs). © 2013 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

Synthetic template: effective tool for target classification and machine vision

A process for replacing a voluminous image dictionary, which characterizes a certain target of interest in a constrained zone of effectiveness representing controlled states including scale and view angle, with a synthetic template has been developed. Synthetic template (ST) is a spatial map (grayscale image) obtained by combining the set of zone-specific training images that are ascribed to the target of interest. It has been shown that the solo-template ST correlation filter outperforms filter banks comprised of multiple target-class training images. A geometric interpretation of the basic ST concept is employed in order to further explain and substantiate its properties.

A Bayesian mixture‐modeling approach for flow‐conditioned multiple‐point statistical facies simulation from uncertain training images

Key Points Preserving prior higher‐order statistics of facies using flow data integration Incorporating the uncertainty in the training image during facies simulation Efficiently sampling from multiple training images based on flow response

Facies Modeling Using a Markov Mesh Model Specification

The spatial continuity of facies is one of the key factors controlling flow in reservoir models. Traditional pixel-based methods such as truncated Gaussian random fields and indicator simulation are based on only two-point statistics, which is insufficient to capture complex facies structures. Current methods for multi-point statistics either lack a consistent statistical model specification or are too computer intensive to be applicable. We propose a Markov mesh model based on generalized linear models for geological facies modeling. The approach defines a consistent statistical model that is facilitated by efficient estimation of model parameters and generation of realizations. Our presentation includes a formulation of the general framework, model specifications in two and three dimensions, and details on how the parameters can be estimated from a training image. We illustrate the method using multiple training images, including binary and trinary images and simulations in two and three dimensions. We also do a thorough comparison to the snesim approach. We find that the current model formulation is applicable for multiple training images and compares favorably to the snesim approach in our test examples. The method is highly memory efficient.

Combining geologic‐process models and geostatistics for conditional simulation of 3‐D subsurface heterogeneity

The goal of simulation of aquifer heterogeneity is to produce a spatial model of the subsurface that represents a system such that it can be used to understand or predict flow and transport processes. Spatial simulation requires incorporation of data and geologic knowledge, as well as representation of uncertainty. Classical geostatistical techniques allow for the conditioning of data and uncertainty assessment, but models often lack geologic realism. Simulation of physical geologic processes of sedimentary deposition and erosion (process‐based modeling) produces detailed, geologically realistic models, but conditioning to local data is limited at best. We present an aquifer modeling methodology that combines geologic‐process models with object‐based, multiple‐point, and variogram‐based geostatistics to produce geologically realistic realizations that incorporate geostatistical uncertainty and can be conditioned to data. First, the geologic features of grain size, or facies, distributions simulated by a process‐based model are analyzed, and the statistics of feature geometry are extracted. Second, the statistics are used to generate multiple realizations of reduced‐dimensional features using an object‐based technique. Third, these realizations are used as multiple alternative training images in multiple‐point geostatistical simulation, a step that can incorporate local data. Last, a variogram‐based geostatistical technique is used to produce conditioned maps of depositional thickness and erosion. Successive realizations of individual strata are generated in depositional order, each dependent on previously simulated geometry, and stacked to produce a fully conditioned three‐dimensional facies model that mimics the architecture of the process‐based model. We demonstrate the approach for a typical subsea depositional complex.

Automatic identification of landmarks in digital images

The authors present an automated system for feature recognition in digital images. Morphometric landmarks are points that can be defined in all specimens and located precisely. They are widely used in shape analysis and a typical shape analysis study involves several hundred digital images. Presently, the extraction of landmarks is usually done manually and the process of identifying the landmarks is an important and labour-intensive part of any such analysis. This process is time-consuming, and quite often the research questions are dependent on the duration of obtaining these data. The authors show that a single training image with its landmark co-ordinates is enough to independently estimate the landmarks of any individual within a particular data set. The reliability and accuracy of the method can be further enhanced by using multiple training images. The precision, repeatability and robustness of the algorithm have been evaluated. It is shown in this study that the method is sufficiently accurate to replace the manual identification of landmarks. The generic nature and intrinsic capability of the feature recognition process enables this method to be easily incorporated into other recognition tasks.

Visual object concept discovery: Observations in congenitally blind children, and a computational approach

Over the course of the first few months of life, our brains accomplish a remarkable feat. They are able to interpret complex visual images so that instead of being just disconnected collections of colors and textures, they become meaningful sets of distinct objects. Exactly how this is accomplished is poorly understood. We approach this problem from both experimental and computational perspectives. On the experimental side, we have launched a new humanitarian and scientific initiative in India, called ‘Project Prakash’. This project involves a systematic study of the development of object-perception skills in children following recovery from congenital blindness. Here, we provide an overview of Project Prakash and also describe a specific study related to the development of face-perception skills following sight recovery. Based in part on the results of these experiments, we then develop a computational framework for addressing the problem of object concept discovery. Our model seeks to find repeated instances of a pattern in multiple training images. The source of complexity lies in the non-normalized nature of the inputs: the pattern is unconstrained in terms of where it can appear in the images, the background is complex and constitutes the overwhelming majority of the image, and the pattern can change significantly from one training instance to another. For the purpose of demonstration, we focus on human faces as the pattern of interest, and describe the sequence of steps through which the model is able to extract a face concept from non-normalized example images. Additionally, we test the model's robustness to degradations in the inputs. This is important to assess the model's congruence with developmental processes in human infancy, or following treatment for extended congenital blindness, when visual acuity is significantly compromised.

Generality of matched filtering and minimum Euclidean distance projection for optical pattern recognition.

Matched filtering followed by a minimum Euclidean distance projection onto realizable filter values was previously shown to optimize the signal-to-noise ratio for single training images in optical correlation pattern recognition. The algorithm is now shown to solve the combination of (1) standard statistical pattern-recognition metrics with multiple training images, (2) additive input noise of known power spectral density and also additive detection noise that is irreducible by the filter, (3) the building of the filter on arbitrary subsets of the complex unit disk, and (4) the use of observable correlator outputs only. The criteria include the Fisher ratio, the Bayes error and Bayes cost, the Chernoff and Bhattacharyya bounds, the population entropy and expected information, versions of signal-to-noise ratio that use other than second power in their norm, and the area under the receiver operating characteristic curve. Different criteria are optimized by different complex scalar weights.