As the years passed by, computers became more powerful and automation became the need of generation. Humans tried to automate their work and replace themselves with machines. This effort of transition from manual to automatic gave rise to various research fields, and document character recognition is one such field. From the last few years, there is a sincere contribution from researchers for the development of optical character recognition systems for various scripts and languages. As a result of intensive research and development, there has been a significant improvement in handwritten devnagari text recognition. The main focus of this study is detailed survey of existing techniques for recognition of offline handwritten Hindi characters. It addresses all the aspects of Hindi character recognition starting from database to various phases of character recognition. The most relevant techniques of preprocessing, feature extraction and classification are discussed in various sections of this study. Moreover, this study is a zest of work accepted and published by research community in recent years. This study benefits its readers by discussing limitations of existing techniques and by providing beneficial directions of research in this field.

Overlapped fingerprints can be potentially present in several civil applications and criminal investigations. Segmentation of overlapped fingerprints is a required step in the process of fingerprint separation and subsequent verification. Overlapped fingerprint segmentation is performed manually (and the resulting manually drawn masks are a required additional input) in all of the overlapped latent fingerprints separation approaches in the literature, which make them only semi-automatic. This study proposes a novel overlapped fingerprint mask segmentation approach, thereby filling that gap in the development of fully automated fingerprint separation solutions. The proposed method uses convolutional neural networks to classify image blocks into three classes – background, single region, and overlapped region. The proposed approach shows satisfactory performance on three different datasets and opens the door for full automation of fingerprint separation algorithms, which is a very promising research area.

Hierarchical graphical models can incorporate jointly several tasks in a unified framework. By applying this approach, information exchange among tasks would improve the results. A hierarchical conditional random field (CRF) is proposed here to improve the semantic image segmentation. Although this newly proposed model applies the information of several tasks, its run time is comparable with the contemporary approaches. This method is evaluated on MSRC dataset and has shown similar or better segmentation accuracy in comparison with models where CRFs or hierarchical models are adopted.

In this study, the authors propose a neural network (NN) method that uses feature points and the angles formed between the points to recognise facial expressions. Accurate facial expression recognition is an important part of affective computing with many practical applications. Yet, achieving acceptable levels of facial recognition accuracy has proven difficult. Feature points and the distances between the points are used to model basic expressions in NN-based approaches, but, in some cases, they cannot generate satisfactory performance. They expand on the characterisation of facial expression by considering the angles formed between feature points to augment the amount of information that is sent to the NNs. Furthermore, to circumvent a common challenge in facial expressions recognition, which is the difficulty of differentiating among several expressions, they designed a post-processing step to assess the output of the NN against a threshold. The whole method makes a decision only when the output of the NN exceeds the threshold. Otherwise, the frame under consideration is assigned to a ‘no decision’ class. They tested our method on the widely used facial expression CK + database and found that it can achieve good accuracy.

This study explores neovascularisation and lesion detection in an integrated framework for gradation in diabetic retinopathy (DR). Imaging is assumed to be done from sub-sample measurements following compressed sensing. Blind estimation of the scale of the matched filter (MF) followed by fuzzy entropy maximisation is done for extraction and classification of the thick and the thin vessels. Mutual information (MI) between vessel density and tortuosity of the thin vessel class is maximised in two dimensions (2D) for neovascularisation detection. For lesion detection, MI between the maximum MF response and the maximum Laplacian of Gaussian filter response is jointly maximised in 2D. The outcomes are then combined in a common platform for gradation in DR. Simulation results demonstrate that 95% images of each of DRIVE, STARE and DIARETDB1 databases and 94% images of MESSIDOR database are correctly graded by the proposed method when 80% measurement space is considered.

This study presents a new fully automated, fast, and accurate brain tumour segmentation method which automatically detects and extracts whole tumours from 3D-MRI. The proposed method is based on a hybrid approach that relies on a brain symmetry analysis method and a combining region-based and boundary-based segmentation methods. The segmentation process consists of three main stages. In the first one, image pre-processing is applied to remove any noise, and to extract the brain from the head image. In the second stage, automated tumour detection is performed. It is based essentially on FBB method using brain symmetry. The obtained result constitutes the automatic initialisation of a deformable model, thus removing the need of selecting the initial region of interest by the user. Finally, the third stage focuses on the application of region growing combined with 3D deformable model based on geodesic level-set to detect the tumour boundaries containing the initial region, computed previously, regardless of its shape and size. The proposed segmentation system has been tested and evaluated on 3D-MRIs of 285 subjects with different tumour types and shapes obtained from BraTS'2017 dataset. The obtained results turn out to be promising and objective as well as close to ground truth data.

In mammograms, the breast skin line often appears ambiguous and poorly defined. This is mainly due to the breast organ compression during the image acquisition process along with the inherent low density of the tissue in that area. The accurate delimitation of the breast region becomes a challenging task to conventional segmentation techniques. In this study, the authors propose a new segmentation approach allowing to overcome this challenge. This approach is based on the application of two complementary segmentation techniques exploring each, respectively, the grey-scale intensities and the local-homogeneity domains. The knowledge resulting from each segmentation technique is considered as a knowledge source and is modelled using the belief functions formalism. The two considered knowledge sources are then fused using an iterative process. The obtained results show the efficiency of the proposed evidential approach especially in terms of ambiguity removal and decision quality improvement for accurate breast border delimitation (which is often under-segmented and assimilated to the background by most of the existing segmentation techniques).

The goal of blind image quality assessment (IQA) is to predict the quality of an image as perceived by human observers without using a reference image. The authors explore a new approach which predicts the image quality based on the conformity of the first digit distribution (FDD) of natural images in the transform domain with Benford's law (BL). The conformity is measured by the symmetric Kullback–Leibler divergence. They first show that while in the transform domain the FDD of a natural image conforms with BL well, the FDD of a distorted natural image violates this conformity. They then train a non-linear regression model which maps features derived from the FDD to the subjective evaluation score of an image. The non-linear mapping is trained using Gaussian process regression with a rational quadratic kernel. The selection of this particular non-linear regression tool is based on extensive experiments and evaluations of many regression tools. They conduct experiments to test the proposed technique using five databases. Results demonstrated that its performance is competitive with those state-of-the-art blind IQA algorithms. In particular, the overall performance of the proposed technique is among the best in all algorithms tested.

The aim of this study is watermarking image encryption based on the fractional Mellin transform (FrMT) and singular value decomposition (SVD) using deterministic phase masks (DPMs). DPMs are used in the input as well as in the frequency planes of double random phase encoding. The use of DPM structured phase mask provides an advantage of extra encryption parameters, besides overcoming the problem of axis alignment associated with an optical setup. The encrypted image resulting from the application of FrMT is attenuated by a factor and then combined with a host image to provide a watermarked image. Afterwards, SVD is performed to get three decomposed matrices, i.e. one diagonal matrix and two unitary matrices. The decryption process is the reverse of encryption. Digital implementation of the proposed scheme has been performed using MATLAB R2014a (8.3.0.532). The watermark image is retrieved by using the corresponding FrMT orders and conjugate of DPMs. Use of the FrMT provides enhanced security due to the non-linear nature of the transform. The effect of noise on the watermarked image has also been investigated. Mean square error between the output and input watermarks shows the accuracy of the proposed scheme.

Texture image analysis plays a pivotal role in pattern recognition and image retrieval. In this study, two improved local binary pattern descriptors are proposed using wavelet transform for texture analysis. The two proposed methods, namely wavelet domain local statistical binary pattern (WLSBP) and directional WLSBP (dWLSBP _α ) both consist of three stages. In the first stage, discrete wavelet decomposition is applied to decompose the image. In the second stage, the proposed statistical parameters are computed from the decomposed image, which results in binary value 0 or 1. Then, the binary values are transformed into WLSBP/dWLSBP _α label. In the third stage, the histogram is built using the WLSBP/dWLSBP _α labels. The proposed WLSBP and dWLSBP _α differ in terms of considering the neighbours. The proposed WLSBP considers the neighbours circularly, whereas dWLSBP _α considers the neighbours in the same orientation through the central wavelet coefficient. The proposed approaches have been applied for copy–move forgery detection. Experiments show that the performance of the proposed methods has improved retrieval rate compared with existing methods on both Brodatz and Outex databases.

This study presents a novel approach for ultrasound (US) images denoising. It concerns a class of generalised method of moments estimators with interesting asymptotic properties for wavelet coefficients 2D generalised autoregressive conditional heteroscedasticity modelling. Afterwards, these estimators can be used for removing noise from US images. Indeed, a minimum mean -square error method is applied for estimating the clean wavelet image coefficients. To judge the quality of the denoising procedure, a link between the denoising efficiency procedure and a proposed asymmetry measure is established. Several tests have been carried out to prove the performance of the proposed approach. The obtained results are compared with those of contemporary image denoising methods using usual image quality assessment metrics and two proposed no-reference quality metrics.

In recent years, convolutional neural networks (CNNs) have been widely used for visual object tracking, especially in combination with correlation filters (CFs). However, the increasing complex CNN models introduce more useless information, which may decrease the tracking performance. This study proposes an online feature map selection method to remove noisy and irrelevant feature maps from different convolutional layers of CNN, which can reduce computation redundancy and improve tracking accuracy. Furthermore, a novel appearance model update strategy, which exploits the feedback from the peak value of response maps, is developed to avoid model corruption. Finally, an extensive evaluation of the proposed method was conducted over OTB-2013 and OTB-2015 datasets, and compared with different kinds of trackers, including deep learning-based trackers and CF-based trackers. The results demonstrate that the proposed method achieves a highly satisfactory performance.

The main task of image segmentation is to partition an image into disjoint sets of pixels called clusters. Spectral clustering algorithm has been developed rapidly in recent years and it has been widely used in image segmentation. The traditional spectral clustering algorithm requires huge amount of computation to process colour images with high resolution. While one possible solution is reducing image resolution, but it will lead to the loss of image information and reduce segmentation performance. To overcome the problem of traditional spectral clustering, an image segmentation algorithm based on superpixel clustering is proposed. Firstly, the algorithm uses the superpixel preprocessing technique to quickly divide the image into a certain number of superpixel regions with specific information. Then, the similarity matrix is used to provide the input information to the spectral clustering algorithm to cluster the superpixel regions and get the final image segmentation results. The experiment results show that the proposed algorithm can effectively improve the performance in image segmentation compared with the traditional spectral clustering algorithm, and finally the substantial improvement has been obtained in respect of computational complexity, processing time and the overall segmentation effect.

Salient object detection is a fundamental problem and has been received a great deal of attention in computer vision. Recently, deep learning model became a powerful tool for image feature extraction. In this study, the authors propose a multi-scale deep neural network (MSDNN) for salient object detection. The proposed model first extracts global high-level features and context information over the whole source image with the recurrent convolutional neural network. Then several stacked deconvolutional layers are adopted to get the multi-scale feature representation and obtain a series of saliency maps. Finally, the authors investigate a fusion convolution module to build a final pixel level saliency map. The proposed model is extensively evaluated on six salient object detection benchmark datasets. Results show that the authors’ deep model significantly outperforms other 12 state-of-the-art approaches.

According to Fresnel's formula and the energy conservation law, a model combining the infrared reflected effect and emitted effect is developed to calculate the polarisation degree. With this model, the polarisation degree difference between the sea surface and ship target in long-wave infrared is simulated. To solve the problem of dim targets detection in a sea background, based on the polarisation difference of the sea surface and ship targets, a method of the non-subsampled shearlets transformation is proposed to fuse the intensity image and polarisation image. The algorithm of distribution coefficients is applied to improve the contrast ratio between targets to background in low-frequency subbands. The denoise scheme of the adaptive threshold is adopted to suppress noise and the conceptions of local direction contrast and region gradient are used as a choosing scheme to the preserve features and edges of images in high-frequency subbands. Image evaluation indices of target contrast with the background and local signal-to-noise ratio are used to evaluate the enhancement effect of fused images. Results show that the evaluation indices of fused images with polarisation features are significantly improved, and comparisons with existing methods demonstrate the effectiveness and reliability of the proposed method.

In this study, a novel technique is proposed to detect the changes in bitemporal multispectral images. Utilisation of the local neighbourhood information in any image processing task may provide good noise immunity and reduces false alarms. Motivated by this, Otsu's thresholding of local information based approach is proposed in this work. It shows the effective performance in change detection of bitemporal Landsat images which suffer from different atmospheric and sunlight conditions. To get the local information around each pixel, both bitemporal images are partitioned into overlapping image blocks. Every block of the first image is concatenated with the corresponding block of the second image for each pixel position. Thus, the information of the concatenated block is considered as inter-block information. Further, Otsu's method is applied on the concatenated block for threshold calculation. Depending on the threshold, binary values are generated. Finally, binary values of both images for all bands are compared by XOR operation to detect it as the background i.e. unchanged pixel or foreground i.e. changed pixel. On the basis of majority class present in XOR output, binary change map is generated. Experiments conducted on Landsat images show that the proposed method provides better performance compared to reported techniques.

Aiming at the high cost and the poor working environment of the detection of large-scale wind turbine (WT) cracks, an analytic detection method based on blade images taken by unmanned aerial vehicles (UAVs) is proposed in this study. For the characteristics of the UAV shooting and the location of the WT, the pre-processing of motion blurring, image noise reduction and image enhancement is used to make the target area and crack details more clear and complete. Then, a crack analysis method based on the grey-scale value is proposed, taking into account the distribution, severity and development trend of the cracks, so that the blind area in the daily detection of the WT can be reduced, the subsequent maintenance of the WT blade is made more accurate, and essentially the operation and maintenance costs be reduced considerably.

Effective texture feature is an essential component in any content-based image retrieval system. In this study, new texture features based on image enhancement technique are presented. The authors have effectively exploited power-law transform (PLT) to extract new spectral texture features called PLT-based spectral features. Extensive experiments on the Brodatz texture database and Salzburg Textures image database prove the effectiveness of the proposed techniques and show that the proposed features significantly outperform the widely used Gabor and curvelet features. The proposed features are also compared with recently published Gaussian copula models of Gabor feature and local tetra patterns (LTrP). The experimental results confirm that the proposed features have more tolerance to scale, orientation and illumination distortion than the state-of-the-art Gabor, curvelet, Gaussian copula models of Gabor and LTrPs.

The automated extraction of retinal blood vessels is the course of action in the medical analysis of retinal diseases. The proposed methodology for the retinal vessel segmentation is based on BAT algorithm and random forest classifier. A feature vector of 40-dimensional including local, phase and morphological features is extracted and the feature set which minimises the classifier error is identified by BAT algorithm. The selected features are also identified as the dominant features in the classification. Performance of the proposed method is analysed by the publicly available databases such as digital retinal images for vessel extraction and structured analysis of the retina. The authors’ proposed method is highly sensitive to identify the blood vessels, in view of the fact that it corresponds to the ability of the method to identify the blood vessels correctly. BAT algorithm-based proposed method achieves very high sensitivity and accuracy of about 82.85 and 95.34%, respectively.

Convolutional neural networks (CNNs) such as encoder–decoder CNNs have increasingly been employed for semantic image segmentation at the pixel-level requiring pixel-level training labels, which are rarely available in real-world scenarios. In practice, weakly annotated training data at the image patch level are often used for pixel-level segmentation tasks, requiring further processing to obtain accurate results, mainly because the translation invariance of the CNN-based inference can turn into an impeding property leading to segmentation results of coarser spatial granularity compared with the original image. However, the inherent uncertainty in the segmented image and its relationships to translation invariance, CNN architecture, and classification scheme has never been analysed from an explicitly spatial perspective. Therefore, the authors propose measures to spatially visualise and assess class decision confidence based on spatially dense CNN predictions, resulting in continuous decision confidence surfaces. They find that such a visual-analytical method contributes to a better understanding of the spatial variability of class score confidence derived from weakly supervised CNN-based classifiers. They exemplify this approach by incorporating decision confidence surfaces into a processing chain for the extraction of human settlement features from historical map documents based on weakly annotated training data using different CNN architectures and classification schemes.

In this study, the problem of detecting if an image has tampered is inquired; especially, the attention has been paid to the case in which the portion of an image is copied and then pasted onto another region to create a duplication or to hide some important portion of the image. The proposed copy-move forgery detection system is based on the scale-invariant feature transform (SIFT) features extraction and density-based clustering algorithm. The extracted SIFT features are matched using the generalised two nearest neighbours (2NN) procedure. Thereafter, the density-based clustering algorithm is utilised to improve the detection results. The proposed system is tested using MICC-F220, MICC-F2000 and MICC-F8multi datasets. Due to the generalised 2NN matching procedure, the proposed system is able to detect multiple forgeries present in the image. Experimental results show that the performance of the system is quite satisfactory in terms of computational time as well as detection accuracy.

Feature extraction from retinal images is gaining popularity worldwide as many pathologies are proved having connections with these features. Automatic detection of these features makes it easier for the specialist ophthalmologists to analyse them without spending exhaustive time to segment them manually. The proposed method automatically detects the optic disc (OD) using histogram-based template matching combined with the maximum sum of vessel information in the retinal image. The OD region is segmented by using the circular Hough transform. For detecting fovea, the retinal image is uniformly divided into three horizontal strips and the strip including the detected OD is selected. Contrast of the horizontal strip containing the OD region is then enhanced using a series of image processing steps. The macula region is first detected in the OD strip using various morphological operations and connected component analysis. The fovea is located inside this detected macular region. The proposed method achieves an OD detection accuracy over 95% upon testing on seven public databases and on our locally developed database, University of Auckland Diabetic Retinopathy database (UoA-DR). The average OD boundary segmentation overlap score, sensitivity and fovea detection accuracy achieved are 0.86, 0.968 and 97.26% respectively.

Recently, in three-dimensional (3D) television, the temporal correlation between consecutive frames of the intermediate view is used together with the inter-view correlation to improve the quality of the synthesised view. However, most temporal methods are based on the motion vector fields (MVFs) calculated by the optical flow or block-based motion estimation which has very high computational complexity. To alleviate this issue, the authors propose a temporal-disparity-based view synthesis (TDVS) method, which uses the MVFs extracted from the bitstreams of side views and motion warping technique to create the temporal correlation between views in the intermediate position. Then a motion compensation technique is used to create a temporal-based view. Finally, the temporal-based view is fused with a disparity-based view which is generated by a traditional depth image-based rendering technique to create the final synthesised view. The fusion of these views is performed based on the side information which is determined and encoded at the sender-side of the 3D video system using a dynamic programming algorithm and rate-distortion optimisation scheme. Experimental results show that the proposed method can achieve the synthesised view with appreciable improvements in comparison with the view synthesis reference software 1D fast (VSRS-1D Fast) for several test sequences.

3D magnetic resonance imaging (3D MRI) is one of the most preferred medical imaging modalities for the analysis of anatomical structures where acquisition of multiple slices along the slice select gradient direction is very common. In 2D multi-slice acquisition, adjacent slices are highly correlated because of very narrow inter-slice gaps. Application of compressed sensing (CS) in MRI significantly reduces traditional MRI scan time due to random undersampling. The authors first propose a fast interpolation technique to estimate missing samples in the k-space of a highly undersampled slice (H-slice) from k-space (s) of neighbouring lightly undersampled slice/s (L-slice). Subsequently, an efficient multislice CS-MRI reconstruction technique based on weighted wavelet forest sparsity, and joint total variation regularisation norms is applied simultaneously on both interpolated H and non-interpolated L-slices. Simulation results show that the proposed CS reconstruction for 3D MRI is not only computationally faster but significant improvements in terms of visual quality and quantitative performance metrics are also achieved compared to the existing methods.

The study of statistical distributions of alternating current (AC) discrete cosine transform (DCT) coefficients is one of the key techniques for digital images. In this study, we have analysed original and power law enhanced images in the logarithmic domain. The logarithmic domain linearises the otherwise nonlinear relation between original and power law enhanced images. We have experimentally proved that Gamma distribution is the best distribution for characterisation of block variance in terms of Jensen–Shannon divergence. Therefore, a composite distribution, Gaussian-Gamma, is employed for characterisation of AC DCT coefficients. We have analytically derived and experimentally verified that the scale parameters of power law enhanced image are proportional to scale parameters of the original image whereas shape parameters remain unchanged. On further experimentation, it is established that scale and shape parameters of the composite statistical distribution of AC DCT coefficients do not change if images are compressed in JPEG format after power law enhancement. Furthermore, a novel feature set of scale parameters is constructed and is applied to train decision tree to classify original, brightened, and darkened images. The comparison of achieved classification results with the state-of-the-art show the efficacy of proposed analysis.