Background subtraction based on change detection is the first step in many video surveillance systems, an effective background subtraction algorithm should distinguish foreground from the background sensitively, and adapt to the variation of background scenes robustly. In this study, the authors propose a robust background subtraction algorithm which takes advantages of local texture features represented by an extended scale invariant local binary pattern and colour intensities to characterise pixel representations. Local texture features achieve good tolerance against illumination variations in rich texture regions but not so efficiently on uniform regions, so a photometric invariant colour measurement is proposed to overcome its limitation. Both quantitative and qualitative evaluations carried out on a well-known change detection dataset are provided to demonstrate the effectiveness of the proposed algorithm.

Automatic optic disc (OD) localisation and segmentation is still a great challenge in computer-aided diagnosis and screening system. Here, a new OD segmentation algorithm is proposed based on the distinct features of OD in terms of its intensity and shape. The algorithm includes four stages: image preprocessing, image segmentation, ellipse fitting, and OD localisation and segmentation. In the preprocessing stage, the blood vessel in the input retinal image is removed by using the morphological operation and median filtering in HSL (hue–saturation–lightness) colour space. In the image segmentation and ellipse fitting stages, the fractional-order Darwinian particle swarm optimisation algorithm is used to extract the brightest region, and the least-squares optimisation is adopted to detect elliptical OD shape. Finally, the smooth OD borders are generated in the last stage. The proposed method is evaluated by the centroid difference, overlapping ratio, overlap score, and success indexes. Experimental results on the retinal images from DRION, MESSIDOR, ORIGA, and many other public databases demonstrate that the proposed method has superior performance, and may be a suitable tool for automated retinal image analysis.

Motivated by the interference of appendages in airline baggage dimension detection using three-dimensional (3D) point cloud, a minimum elastic bounding box (MEBB) algorithm for dimension detection of 3D objects is developed. The baggage dimension measurements using traditional bounding box method or shape fitting method can cause large measurements due to the interference of appendages. Starting from the idea of ‘enclosing’, an elastic bounding box model with the deformable surface is established. On the basis of using principal component analysis to obtain the main direction of the bounding box, the elastic rules for deformable surfaces are developed so as to produce a large elastic force when it comes into contact with the main body part and to produce a small elastic force when it comes into contact with the appendages part. The airline baggage measurement shows how to use MEBB for dimension detection, especially for the processing of isotropic density distribution, the elasticity computing and the adaptive adjustment of elasticity. Results on typical baggage samples, comparisons to other methods, and error distribution experiments with different algorithm parameters show that the authors’ method can reliably obtain the size of the main body part of the object under the interference of appendages.

Magnetic resonance imagings (MRIs) have different modalities, including T1- and T2-weighted, PD (proton density), and Flair images. Brain MRI segmentation is a challenge when coping with artefacts such as intensity non-uniformity, partial volume effects, and noise. As artefacts change the intensity of different part of MRI modalities, describing the intensity of these modalities is highly uncertain. Here, it is proposed that the Dempster–Shafer theory and fuzzy clustering can be combined for brain MRI segmentation because of their robustness. The purpose of this research is to offer a technique based on data fusion of different modalities to segment brain MRIs. T1, T2, PD, and Flair were employed in this study. Fuzzy clustering and specific mapping were used to form the Dempster–Shafer belief structure. In order to evaluate the efficiency of the proposed algorithm, several simulations were performed and the Dice and Jaccard coefficients were used to compare the results with those of other methods. The qualitative and quantitative results of the proposed algorithm verify the success of the proposed algorithm. This method improved 3–4% over that of the previous methods which had showed the best results.

The authors propose an image region annotation framework by exploring syntactic and semantic correlations among segmented regions in an image. A texture-enhanced image segmentation JSEG algorithm is first used to improve the pixel consistency in a segmented image region. Next, each region is represented by a set of image codewords, also known as visual alphabets, with each of them used to characterise certain low-level image features. A visual lexicon, with its vocabulary items defined as either a codeword or a co-occurrence of multiple alphabets, is formed and used to model middle-level semantic concepts. The concept classification models are trained by a maximal figure-of-merit algorithm with a collection of training images with multiple correlations, including spatial, syntactic and semantic relationship, between regions and their corresponding concepts. In addition, a region-semantic correlation model constructed with latent semantic analysis is used to correct the potentially wrong annotations by analysing the relationship between image region positions and labels. When evaluated on the Corel 5K dataset, the proposed image region annotation framework achieves accurate results on image region concept tagging as well as whole image based annotations.

Breast ultrasound (BUS) images are of poor quality, contain inherent noise and shadow regions. Consequently, the task of tumour segmentation from these images becomes more difficult. In this study, a modified spatial neutrosophic clustering technique has been proposed for automatic boundary extraction of tumours in B-mode BUS images. The contributions of the work are two-fold: (i) spatial information is incorporated in the neutrosophic ℓ-means (NLM) clustering method for better cluster formation and (ii) membership values are updated by using type-2 membership function, which helps in converging the cluster centres to more desirable locations than ordinary fuzzy membership functions. BUS images with manually marked lesions by an experienced radiologist have been used as gold standard/reference images for quantitative comparison. The proposed method has been applied to 60 BUS images and results are recorded in the form of area and boundary error metrics. The performance of the proposed method has been compared with the region growing, fuzzy c-means clustering, watershed segmentation, neutrosophic c-means clustering and NLM clustering methods. From the quantitative and visual results, it has been observed that the proposed method can extract the tumour boundaries more precisely as compared with the other state-of-the-art clustering techniques.

To obtain a higher accuracy in the multi-atlas patch-based label fusion method, it is essential to have the accurate similarity measure of selected patches. In this study, the authors propose a new sparse patch-based representation method using a local binary texture (LBT) in the atlas image and atlas label information for the multi-atlas label fusion. In the proposed method, the intensity information in a patch is converted into a LBT which is then combined with the labels of corresponding patches from the atlas to form an atom of a dictionary. The initial labels of target images are estimated through a rough segmentation. The voxel in a patch to be labelled is also constructed as a vector similar to the atom. The voxel vector is then modelled as a sparse linear combination of the atoms in the dictionary. Experimental results on two MR brain data sets demonstrated that the proposed method is efficient in the segmentation which can achieve competitive performance compared with the state-of-the-art methods.

The complex wavelet transforms (CWTs) are known for their excellent edge preserving together with nearly shift invariant features. This study presents new orthogonal CWT filter designs. The proposed designs guarantee the perfect reconstruction feature of the CWT system while satisfying in a least squares sense, the CWT Hilbert constraints over the filter's pass-band. The study also describes a new image denoising technique based on bivariate thresholding of the noisy CWT coefficients. In this respect, a simple model is proposed to model the dependence between the magnitudes of the children and parent wavelet coefficient at every sub-band. This allows the derivation of a closed-form expression for the clean thresholded magnitudes. Several illustrative examples are given to verify the superior denoising performance especially in salt & pepper case.

Traditional image stitch methods based on feature point detection require long registration time for high resolution images. In the SIFT (scale invariant feature transform) algorithm, it builds difference-of-Gaussian by a linear Gauss expansion filter to obtain the feature. SIFT has poor real time. In this study, a novel image registration method based on image block is proposed to make a rough match for the blocked image and fine match in the most similar blocks by taking advantage of the FAST (features from accelerated segment test) algorithm which runs faster. This way can avoid spending a lot of time in the ineffective area, and enhance the precision and efficiency of the feature point. The accumulative error exists in the process of image stitching, so the image stitched by multiple images has wavelike effects, tilt, or distortion. The camera calibration method is utilised to eliminate the tilt and distortion of the image. The algorithm combining the optimal seam and multi-resolution fusion is adopted to fuse the stitched image and realise seamless stitch of multiple images in order to achieve a seamless image of high resolution. Simulation experimental results show that the stitching method could realise seamless stitching and straightening of multiple images.

Proportions of different phases (phase fraction) in the microstructures determine the quality of dual phase (DP) steel. So, calculation of phase fraction in the microstructures of steel samples is important for quality assurance. Manual calculation of phase fraction involves Le Pera etching of steel which is time consuming and dependent on operator efficiency. Calculation of phase fraction from Le Pera etched samples requires cumbersome manual observations. Nital etching is a faster alternative to Le Pera etching. However, due to lack of visually discriminative information, different phases cannot be identified manually from nital images. We propose a novel method for automatic calculation of phase fractions in steel microstructures from nital images using machine learning techniques. We show that regional contour patterns and local entropy (which cannot be evaluated manually) of regions of nital images are related to the formation process of the phases. We design a method that automatically evaluates regional contour patterns and local entropy from nital images of DP steel. Subsequently, we construct a random forest classifier that uses regional contour patterns and local entropy as features for classification of different phases. Our method is ∼150 times faster than manual classification. Experiments show close to 90% accuracy in classification.

A fast adaptive and selective mean filter is presented to remove salt and pepper noise effectively from images corrupted with higher noise densities. The algorithm achieves better results in terms of visual quality and in terms of peak signal-to-noise ratio, mean absolute error, mean structural similarity index measure, image enhancement factor, and edge preservation ratio than many existing state-of-the-art algorithms at all noise densities. Adaptive filters that use variable window size produce better restoration of salt and pepper noise at higher noise densities than filters that use fixed window size, but they consume more time. This makes them practically impossible to implement them in digital image acquisition devices. Hence, reducing the execution time of adaptive filters is vital. The proposed algorithm consumes around 90% less time for lower noise densities and 50% less time for higher noise densities than the adaptive weighted mean filter, one of the best available adaptive filters in the literature for high-density salt and pepper noise removal.

Texture classification is one of the recently popular study topics in pattern recognition. Local binary pattern (LBP) is a very efficient local texture descriptor and is used for feature extraction in texture recognition. There are five main steps in representation of texture images: neighbourhood topology and sampling, thresholding and quantisation, encoding and regrouping, combining complementary features. In this study, the authors used symmetric two spirals LBP to measure the grey-scale difference between the centre pixel and its neighbours. They also extended the proposed method by using four spirals LBP to generate the LBP code. For classification, linear regression classification method, which is generally used to solve the face recognition problems, is used. The authors tested the performance of their method on UIUC and CUReT texture image databases. It is experimentally demonstrated that the proposed method achieves the highest classification accuracy among the comparative methods on texture databases.

In image denoising, high-frequency components are more notable to the human eyes than low-frequency components. While high-frequency components contain more variations and represent the detailed textures, the reconstructions of these components are much harder and it is a remaining challenge in image denoising. In this study, a novel edge-preserving image denoising algorithm is proposed, it treats the low- and high-frequency components of the image separately. For restoration of high-frequency components, a neighbourhood regression method is proposed. An energy minimisation function is developed to combine the low- and high-frequency components into one model. Experiments show that the proposed method outperforms the state-of-the-art methods in peak signal-to-noise ratio, edges preservation and visual performance.

In traditional image fusion, source images are separated into a fixed space. The low-frequency part and the high-frequency part are not discriminated according to the nature of the image. Traditional fusion rules often use a fixed proportion, causing colour distortion. In this study, a new adaptive decomposition algorithm is proposed to distinguish high frequency and low frequency of structure image to obtain smoothing layer and texture layer. The smoothing layer of the structural image and the colour information of the function image are fused according to dynamic rules, and then the texture layer is added. On the basis of the objective evaluation metrics, the spectral information evaluation metrics are introduced to evaluate the retention of colour. In the experiments, the proposed method is compared with other six classical image fusion methods. The experiment results show that the proposed method can retain the colour information and structure information very well at the same time. Concerning subjective and objective evaluation, the proposed algorithm is superior to other algorithms.

Researchers utilised various types of information in active contour models to define new energy functionals for image segmentation. These models aim to extract all potential objects from the background, but non-target objects and noise are also obtained. In this study, the authors aim to extract target objects with sparse representation method. The original indicator function (a binary function) with respect to the level set function is used to represent the foreground (value is 1) and background(value is 0). From another point of view, an indicator function can be represented by linear combination of a set of the basis function. Firstly, by a label operator for the indicator function in each iteration, every connected area is represented by a basis function. Secondly, the linear combination of these basis functions is used to represent objects. Finally, through the sparsity constraint of coefficients of basis functions, the object extraction is viewed as a sparse representation problem. Meanwhile, a corresponding improved orthogonal matching pursuit algorithm is designed to obtain the ideal results. Experiments demonstrate that the proposed method has superior performance in object extraction compared with state-of-the-art active contour models. Furthermore, the proposed method also increases the flexibility of applications.

Visual object tracking is an important and challenging task in computer vision. In this study, the authors propose a novel visual tracking approach by decomposing the tracking task into translation and scale estimation. In translation estimation, they employ multiple adaptive correlation filters with features of hierarchical convolutional neural networks (CNNs) to more accurately estimate the target location. To make full use of multi-level features from different CNN layers, they propose an adaptive weighted algorithm to fuse correlation response maps. In scale estimation, a one-dimensional correlation filter with histogram of oriented gradient (HOG) features is employed to estimate the scale variation. Extensive experimental results on 50 challenging benchmark video sequences demonstrate that the proposed algorithm outperforms state-of-the-art algorithms.

Smooth filtering is a common post-operation which is exploited to blur and conceal the traces of tampered objects. Most of the existing forensic methods aim at detecting only one type of filtering process, such as median filtering or Gaussian filtering, which limits their applications. The authors present a new forensic method based on deep learning technique, which utilises a convolutional neural network (CNN) to automatically learn hierarchical representations from the input images. Unlike conventional CNN models, a modified CNN architecture is specifically designed to identify traces left by the manipulation. A filter layer is added into the CNN. The filtering residual in frequency feature of the input image is extracted by this added layer. The output feature is then fed into the next layer of the CNN. Radon transform is applied to increase the distinctiveness of the residual feature. Experimental results on several public datasets show that the proposed CNN-based model outperforms some state-of-the-art methods.

Forgery detection from the images is gaining remarkable interest as there are a lot of editing tools that enable to cause edition with manipulation or removal of the objects from the images. This study proposes a new forgery detection scheme that is based on the supervised learning approach. The supervised learning is brought about by using the support vector neural network and the optimisation is enabled using the fruit fly optimisation algorithm. Initially, the images are fed to the texture descriptor and the face is detected using the Viola–Jones algorithm. The face detected images are subjected to the feature extraction using the Gabor filter + wavelet + texture operator and the features are concatenated to present the input to the classifier. Then, the classifier which is trained using the fruit fly optimisation classifies the features to detect the presence of the manipulation. The performance of the proposed scheme is evaluated with the existing methods for the evaluation metrics accuracy, sensitivity, and specificity using two datasets, namely DSO-1 and DSI-1. The analysis shows that the proposed scheme attained an accuracy of 0.9523, the sensitivity of 0.94, and the specificity of 0.9583, which are greater when compared to the existing methods.

Changes of retinal blood vessel are precursors of many serious diseases such as diabetic retinopathy, hypertension and cardiovascular diseases. Automatic segmentation of retinal blood vessels in the fundus image can better assist in the diagnosis of these diseases and has been studied by many researchers. However, the segmentation of pale vessel pixels remains a problem because of their low contrasts with surrounding pixels. This study proposes an improved multi-scale line detector to segment retinal vessels. It computes the line responses of vessels in multi-scale windows and takes the maximum as the response value, which can enhance the responses of pale vessel pixels near strong vessels or dark background pixels. Experimental results on the publicly available database DRIVE demonstrate that the proposed method can detect pale vessel pixels better. It achieves 75.28% in sensitivity and 94.47% in accuracy, which outperforms the state-of-the-art unsupervised methods. Compared with the supervised methods it also gets better sensitivity and comparable accuracy.

In this study, a thorough analysis of image reconstruction in an active millimetre-wave (MMW) cylindrical imaging system is performed. To improve the system's performance in terms of the total cost as well as the coupling effect, sparse multi-static arrays are desired. Nevertheless, the existing reconstruction methods fail to operate in sparse multi-static antenna-array systems. Therefore, to address these shortcomings, in this study, at first, the existing MMW mono-static cylindrical image reconstruction methods are modified in both single-frequency and wideband reconstruction platforms. Moreover, in order to further improve the image quality and the system cost, a sparse multi-static one-dimensional antenna array has been proposed. To use this sparse array in the cylindrical MMW imaging system, a novel multi-static cylindrical image reconstruction method is also proposed. The performances of the proposed reconstruction methods for both mono-static and sparse multi-static cylindrical imaging structures are validated through numerical simulations. Through the obtained results, 83% reduction in antenna array cost along with a 25 dB decrease in coupling effect is achieved at the cost of a minor reduction in the reconstructed image quality.

This study presents a novel ensemble classifier-based off-line handwritten word recognition system following a holistic approach. Here each handwritten word is recognised using two handcrafted features, namely (i) Arnold transform-based feature that addresses local directional feature which depends on the stroke orientation distribution of cursive word and (ii) oriented curvature-based feature which is basically the histogram of curvelet index and one machine generated feature using deep convolution neural network (DCNN). In this study, a new architecture of DCNN is proposed for handwritten word recognition. These features are recognised by three classifiers separately. Finally, the decision of three classifiers is combined to predict the ultimate word class level. To fuse the decision of individual classifiers, the authors have explored three strategies: (i) vote for strongest decision, (ii) vote for majority decision and (iii) vote for the sum of the decisions. The proposed handwritten word recognition system is tested on three handwritten word databases: (i) CENPARMI database, (ii) IAM database and (iii) ISIHWD database. The performance of the proposed system is promising and comparable to state-of-the-art handwriting recognition systems.

This work proposes a variable neighbourhood search (FTS) that uses a fractal-based local search primarily designed for images. Searching for specific content in images is posed as an optimisation problem, where evidence elements are expected to be present. Evidence elements improve the odds of finding the desired content and are closely associated to it in terms of spatial location. The proposed local search algorithm follows the fashion of a chain of triangles that engulf each other and grow indefinitely in a fractal fashion, while their orientation varies in each iteration. The authors carried out an extensive set of experiments, which confirmed that FTS outperforms state-of-the-art metaheuristics. On average, FTS was able to locate content faster, visiting less incorrect image locations. In the first group of experiments, FTS was faster in seven out of nine cases, being >8% faster on average, when compared to the second best search method. In the second group, FTS was faster in six out of seven cases, and it was >22% faster on average when compared to the approach ranked second best. FTS tends to outperform other metaheuristics substantially as the size of the image increases.

We present two variants of a colour space transformation algorithm to encode Bayer colour filter array images that are based on integer coefficients; as a result, the algorithms are fully lossless and reversible in nature. These transformation algorithms are derived using an optimisation model that reduces the spectral redundancy of Bayer colour components, which results in lower prediction error variance and inter-colour correlation. These methods, known as optimum reversible colour space transform (ORCT-1 and ORCT-2), improve the lossless bitrate of low complexity prediction model without using high complexity interpolation and inter-colour prediction scheme. Extensive experimentation is performed using five sets of test images for different lossless compression algorithms: JPEG-LS, JPEG-2000 and JPEG-XR. Experimental results show that, in all cases, the proposed schemes perform competitively with other methods with lower computational complexity, which makes them suitable for low-cost imaging applications.

Histopathology image segmentation is an important area in the field of computer aided diagnosis using image processing. This study presents a local feature-based novel technique for the segmentation of histopathology images. It mainly focuses on a system that segments overlapped nuclei (OLN) without affecting the general non-OLN segmentation performance. The proposed method suggests a three-stage system. The initial segmentation is done by using local features for the demarcation of nuclei regions. In the second stage, salient-based active contour is applied for complete nucleus-region identification. In the final step, the OLN are identified and segmented using a Gaussian distribution and entropy maximisation. The performance of the proposed segmentation method is evaluated on the basis of precision, recall, accuracy, and -score. The proposed method is simulated on animal diagnostics laboratory histopathology image dataset and reported 90.3% average accuracy with average F ₁-score 0.937. Simulation results confirm the superiority of the proposed method as compared with the existing state-of-art methods.

Correlation-based pattern recognition filtering methods such as the eigenextended maximum average correlation height (EEMACH) filter is considered an effective tool in object recognition applications. However, these approaches require exclusive training for all possible distortions including in-plane as well as out-of-plane rotation, scale and illumination variations. The overall training process is exhaustive and requires training of filter banks to handle specific types of distortion separately. To overcome the aforementioned limitations, the authors propose a new difference of Gaussian (DoG)-based logarithmically preprocessed EEMACH filter which can manage multiple distortions in a single training instance while ensuring inherent control over illumination variations. The DoG-based logarithmic treatment exploits inherent capabilities of logarithmic preprocessing to manage scale and in-plane rotations. By reducing the number of classifier instances to one third, it not only reduces the computation complexity of the process to 33%, but also enhances the object recognition performance. The cumulative improvement is up to 2.73% in case of rotations and 10.8% in case of scaling by incorporating reinforced edges due to DoG operation. The resultant filter displays significantly enhanced recognition performance leading to a higher percentage of correct machine decisions, especially when an input scene contains multiple distortions.

The following article published in IET Image Processing, Shahid, Muhammad; Taj, Imtiaz A.: ‘Robust retinal vessel segmentation using vessels location map and Frangi enhancement filter’, IET Image Processing, 2018, DOI: 10.1049/iet-ipr.2017.0457 on 16th January 2018 has been retracted due to a breach of the IET's Policy in Relation to Plagiarism, Infringement of Copyright and Infringement of Moral Rights and Submission to Multiple Publications. Prof. Imtiaz Ahmed Taj was unaware of and not complicit in any misconduct.