Recently, image de-noising algorithm based on sparse representation has received an increasing amount of attention. Such algorithms proposed a comprehensive sparse representation model, by solving the sparse coding problem and choosing the proper method for dictionary updating to achieve better de-noising results. Therefore, the construction of learning dictionary has become one of the key problems that limit the de-noising effectiveness. The non-locally centralised sparse representation de-noising algorithm uses principal component analysis method to achieve dictionary updating. Nevertheless, the instability of a single complete dictionary in sparse coding leads to erratic result in the process of the original image restoration. In this study, the authors present a new method named generalised non-locally centralised sparse representation algorithm. In the proposed method, the authors cluster the training patches extracted from a set of example images into subspaces, and then train dictionaries for subspaces by sparse analysis k-singular value decomposition dictionary, which is utilised to construct coded sub-block dictionary to avoid the instable results caused by a single dictionary. Experiments show that the improved method has better signal-to-noise ratio and de-noising effect compared with other methods.

In this study, the authors propose a new method to fuse multiple segmentations generated by different methods or same methods with different parameters. The proposed method has several contributions. First, they convert the image segmentation fusion problem into a weakly supervised learning problem. Thus, the information of superpixels can be used to guide the fusion process. Second, they treat the multiple segmentations as multiple closely related tasks and utilise multi-task learning method to evaluate the reliability of the segmentations. Third, they design a strategy to ensemble the evaluated segmentation maps to obtain the final segmentation. The experiment on the benchmark dataset MSRC demonstrates the superior performance of the proposed method on image foreground and background segmentations.

Image filtering is to retain the details of the image as much as possible and meanwhile suppress the noise pollution to great extent. This study presents an image filtering using the truncated statistics and edge preserving. In the first step of our method, the alpha-trimmed filter is utilized to remove a variety of types of noises; in the second step, taking the image after alpha-trimmed filtering as a guide image, the local linear model between the guide image and the target image is established; in the third step, the obtained local linear model is further simplified to reduce the time complexity; and finally, using the relationship between image local variance and the global variance, the local linear model is modified to enhance the details of the image and meanwhile remove halo phenomenon. This method has three advantages: (i) it is flexible to deal with the images stained by various types of high-intensity noise; (ii) it is effective to keep the image details and profile information, and remove the halo phenomenon; and (iii) it runs in time linear in the image size, thus its computation complexity is low. Experimental results show that the proposed filter is robust and efficient.

The possibility theory, which is an extension of fuzzy sets and fuzzy logic, has shown considerable potential for solving active learning (AL) problems, particularly for multiclass scenarios’ classification. Hence, two recently proposed fuzzy multiclass AL algorithms (classification ambiguity (CA) and fuzzy C-order ambiguity (FCOA)) are investigated to properly generalise them for classifying hyperspectral images, and two improved versions of the CA and FCOA are proposed. In addition to comparing the performances of the original and improved algorithms, several other state-of-the-art AL methods are evaluated, such as breaking ties, margin sampling, and multi-class level uncertainty, with or without diversity criteria such as angle-based diversity (ABD), clustering-based diversity (CBD), and enhanced clustering-based diversity (ECBD). Tests on two benchmark hyperspectral images confirm that the proposed improved algorithms are superior to and more effective than the original ones.

Automatic image classification has become a necessary task to handle the rapidly growing digital image usage. It has branched out many algorithms and adopted new techniques. Among them, feature fusion-based image classification methods rely on hand-crafted features traditionally. However, it has been proven that the bottleneck features extracted through pre-trained convolutional neural networks (CNNs) can improve the classification accuracy. Thence, this study analyses the effect of fusing such cues from multiple architectures without being tied to any hand-crafted features. First, the CNN features are extracted from three different pre-trained models, namely AlexNet, VGG-16, and Inception-V3. Then, a generalised feature space is formed by employing principal component reconstruction and energy-level normalisation, where the features from individual CNN are mapped into a common subspace and embedded using arithmetic rules to construct fused feature vectors (FFVs). This transformation play a vital role in creating a representation that is appearance invariant by capturing complementary information of different high-level features. Finally, a multi-class linear support vector machine is trained. The experimental results demonstrate that such multi-modal CNN feature fusion is well suited for image/object classification tasks, but surprisingly it has not been explored so far by the computer vision research community extensively.

A new linear colour image filter based on linear quaternion systems (LQSs) is introduced. It detects horizontal, vertical, left- and right-diagonal edges with a single LQS convolution mask. The proposed filter is a canonic minimal filter of four LQS filters, each with different angles of rotation combined parallel wise. Different angles of rotation are a key features of the new filter such that horizontal, vertical, left, and right-diagonal LQS filter masks rotate pixels through angles , , , and , respectively. Although, the four LQS masks are combined parallel to make a single LQS mask but derived using four quaternion convolutions, one for each direction of edges, the LQS filter produces a result without the combination of results from four separate edge detectors. This methodology could be generalised to design more elaborate LQS filters to perform other geometric operations on colour image pixels. The proposed filter translates smoothly changing colours to different shades of grey and produces coloured edges in multiple directions, where there is a sudden change of colour in the original image. Another key idea of the proposed filter is that it is linear because it operates in homogeneous coordinates.

In this paper, a novel method of active contours based on the formulation of partial differential equation (PDE) is proposed for image segmentation. The evolution equation incorporates a force term that pushes the contour towards object boundary, a regularisation term which takes into account the smoothness of the level set function and an edge term which helps to stop the contour at required boundaries. The proposed method integrates an image convolved by a variable kernel into an energy formulation, where the width of the kernel varies in each iteration. Therefore, it takes local region information when the width of the kernel is small while for the larger width of the kernel, the proposed method considers global region information across the regions. Due to the use of both local and global image information, the method easily detects objects in the complex background and also segments the objects where intensity changes within the object. Moreover, the proposed method totally eliminates the need of the contour initialisation by using constant initialisation scheme. Experimental results on real and medical images prove the robustness of the proposed method. Finally, the authors validate their method on PH2 database for skin lesion segmentation.

Combining the advantage of Contourlet transform with adaptive fuzzy Markov random field (MRF) model, a novel segmentation algorithm is proposed in this study to achieve precise and continuous division for synthetic aperture radar (SAR) image. The classical MRF model is modified to obtain the edge and process adaptive label factor. The adaptive fuzzy MRF (AFMRF) model is proposed to equilibrate the smooth regions segmentation with texture regions segmentation. At the same time, Contourlet domain hidden Markov tree (HMT) model is introduced to perform multi-scale directional filtering and intra-scale training on the coefficients of SAR images to achieve precise texture segmentation at each scale. Finally, the AFMRF model is integrated into inter-scale and intra-scale HMT training results and the segmented image can be obtained. To verify the validity of the proposed algorithm, several SAR images are experimented and compared with the state-of-the-art algorithms. The experimental results and analysis show that the proposed algorithm can achieve better results on noise suppression, smoothness of target regions, precise and continuous segmentation of fuzzy texture.

The authors propose an automatic method for extracting objects with fine quality from photographs. The authors’ method starts with finding bounding boxes that enclose potential objects, which is achievable by state-of-the-art object proposal methods. To further segment objects within obtained bounding boxes, the authors propose a new multi-pass level-set method based on saliency detection and foreground pixel classification. The level-set function is initially constructed with respect to the automatically detected salient parts within the bounding box, which eliminates potential user interaction and predicts an initial set of pixels on the object. The input features for foreground pixel classifiers are constructed as a combination of classical texture features from the Gabor filter banks and convolutional features from a pre-trained deep neural network. Through multi-pass evolution of the level-set function and re-training of the foreground pixel classifier, the authors’ method is able to overcome possible inaccuracies in the initial level-set function and converge to the real object boundary.

Under low bit-rate requirements, JPEG baseline causes compression artifacts in the image. Through this paper, a novel region-of-interest (ROI) dependent quantization method in JPEG framework is proposed. The proposed method judiciously quantizes DCT coefficients belonging to salient and non-salient regions of the image. In this work, multiple ROIs are optimally identified and ranked by using variances. The number of classes is adaptively calculated using goodness-of-segmentation. After the number of classes and their ranks are obtained, the image is divided into blocks of size . These blocks may belong to more than one class and hence these are ranked based on their membership in various classes. 2D-DCT coefficients of each block are obtained and then quantized adaptively based on their ranks. Overhead for rank information of blocks is minimized by applying delta-encoding. Results are analyzed in terms of objective quality parameters and visual perception and found that the blocking artifacts in the proposed method are significantly lower than JPEG. The efficiency of the proposed method is demonstrated by results of recently published similar methods and the former is found superior in terms of quality of the reconstructed image.

The domain of noise fading from digital images, by virtue of its enormous appellation amongst the researchers, stands out uniquely in the recent research field of image processing over the last few decades. Periodic noises are unintended spurious signals which often agitate an image during acquisition/transmission, thereby resulting in repetitive patterns having spatial dependency and extensively demeaning visual excellence of the image. However, high amplitude noisy spectral components are clearly noticeable from the remaining uncorrupted ones in the corresponding Fourier transformed corrupted image spectrum. Hence, it is easier to distinguish and minimise those noisy components using an appropriate thresholding and filtration technique. Therefore, to start with, a simple yet elegant model of the noise-free natural image has been developed from the corrupted one followed by a proper thresholding method to get the noisy bitmap. Finally, an elegant adaptive sinc restoration filter with the concept of extracting the exact shape of a noise spectrum profile has been applied in the filtration phase. The performance of the proposed algorithm has been assessed both visually and statistically with other state-of-the-art algorithms in the literature in terms of various performance measurement attributes, providing evidence of achieving more effective restoration with considerable lower computational time.

Here, the authors explore the potential of multispectral imaging applied to image completion. Snapshot multispectral cameras correspond to breakthrough technologies that are suitable for everyday use. Therefore, they correspond to an interesting alternative to digital cameras. In their experiments, multispectral images are acquired using an ultracompact snapshot camera-recorder that senses 16 different spectral channels in the visible spectrum. Direct exploitation of completion algorithms by extension of the spectral channels exhibits only minimum enhancement. A dedicated method that consists in a prior segmentation of the scene has been developed to address this issue. The segmentation derives from an analysis of the spectral data and is employed to constrain research area of exemplar-based completion algorithms. The full processing chain takes benefit from standard methods that were developed by both hyperspectral imaging and computer vision communities. Results indicate that image completion constrained by spectral presegmentation ensures better consideration of the surrounding materials and simultaneously improves rendering consistency, in particular for completion of flat regions that present no clear gradients and little structure variance. The authors validate their method with a perceptual evaluation based on 20 volunteers. This study shows for the first time the potential of multispectral imaging applied to image completion.

In this study, a multiple hypothesis tracking (MHT) algorithm for multi-target multi-camera tracking (MCT) with disjoint views is proposed. The authors' method forms track-hypothesis trees, and each branch of them represents a multi-camera track of a target that may move within a camera as well as move across cameras. Furthermore, multi-target tracking within a camera is performed simultaneously with the tree formation by manipulating a status of each track hypothesis. Each status represents three different stages of a multi-camera track: tracking, searching, and end-of-track. The tracking status means targets are tracked by a single camera tracker. In the searching status, the disappeared targets are examined if they reappear in other cameras. The end-of-track status does the target exited the camera network due to its lengthy invisibility. These three status assists MHT to form the track-hypothesis trees for multi-camera tracking. Furthermore, a gating technique which eliminates the unlikely observation-to-track association using space-time information has been introduced. In the experiments, the proposed method has been tested using two datasets, DukeMTMC and NLPR\_MCT, which demonstrates that the method outperforms the state-of-the-art method in terms of improvement of the accuracy. In addition, real-time and online performance of proposed method is also showed in this study.

Since the division with remainder cannot be implemented in multivariable polynomials, the two-dimensional non-separable wavelet transform cannot be lifted by using a similar way as that of univariate wavelet transforms. To solve this problem, a general lifting factoring method of two-dimensional two-channel non-separable stripe filter banks is presented. The constructing form of the polyphase matrices of the stripe filter banks is deduced and the general factoring of the polyphase matrices is given. Compared with the separable lifting wavelet transform, the proposed lifting factoring method can extract better texture information. The lifting form is more succinct than that of the tensor product lifting wavelet transform. The computation amount of the proposed factoring method for image decomposition is a quarter of the two-dimensional two-channel non-separable stripe filter bank and the original two-dimensional two-channel non-separable wavelet system is quickened. Moreover, the proposed lifting factorising method is faster than the traditional two-dimensional two-channel non-separable wavelet transform based on the Fourier transformation framework in which the size of each filter is greater than . The proposed lifting factorising method has better sparsity than that of the original wavelet transform and the famous two-dimensional two-channel biorthogonal symmetric non-separable wavelet transform.

This study proposes two reversible data hiding (RDH) schemes for vector quantisation (VQ)-compressed images based on switching-tree coding (STC) and dynamic tree-coding scheme (DTCS). Most developed VQ-based RDH schemes produce non-legitimate codes as output. In order to preserve the legitimacy of the embedded VQ code, some schemes embed data into VQ indices by employing an index replacement mechanism and some other schemes perform embedding by adopting one of the possible ways during encoding each index when multiple ways are possible to encode the index. In the current research, two schemes are proposed based on the second mechanism. Outputted code of the proposed schemes is a legitimate STC/DTCS code and the conventional STC/DTCS decoder can decode it to the original VQ index table. The experimental results show that the proposed schemes are feasible and in comparison with some previous RDH schemes, the first one provides higher embedding capacity and the second one embeds a substantial amount of data while provides lower bit rate than most the previous schemes. In addition, the embedding-efficiency of both proposed schemes is higher than that of the previous schemes.

Source camera identification (SCI) is a forensic problem of mapping an image back to its source, often in relation to cybercrime. In this digital era, this problem needs to be addressed with the utmost care as a falsely identified source might implicate an innocent person. A very practical problem in this study is the presence of unknown models in the set of cameras under question. In other words, the images under question might not have originated from any of the camera models that are accessible to the forensic analyst, but from a different inaccessible source. Under such a circumstance, the conventional source detection techniques fail to identify the correct source, and falsely map the image to one of the accessible camera models. To address this problem, here the authors propose an SCI scheme which is capable of identifying N known (accessible) as well as K unknown (inaccessible) camera models. The authors’ experimental results prove that the proposed scheme efficiently separates the known and unknown models, and helps to achieve considerably high source identification accuracy as compared to the state-of-the-art.

The authors present a novel online visual tracking algorithm via ensemble autoencoder (AE). In contrast to other existing deep model based trackers, the proposed algorithm is based on the theory that the image resolution has an influence on vision procedures. When the authors employ a deep neural network to represent the object, the resolution is corresponding to the network size. The authors apply a small network to represent the pattern in a relatively lower resolution and search the object in a relatively larger area of the neighbourhood. After roughly estimating the location of the object, the authors apply a large network, which can provide more detailed information, to estimate the state of the object more accurately. Thus, the authors employ a small AE mainly for position searching and a larger one mainly for scale estimating. When tracking an object, the two networks interact to operate under the framework of particle filtering. Extensive experiments on the benchmark dataset show that the proposed algorithm performs favourably compared with some state-of-the-art methods.

A novel impulse noise filter that preserves the image details and effectively suppresses high-density noise has been proposed in this work. The proposed filter works in two phases: (i) noise pixel detection phase and (ii) noise pixel restoration phase. In the detection phase, the impulse noise corrupted pixels are detected using a neighbourhood decision approach. In the second phase, the true values of corrupted pixels are restored using a first-order neighbourhood decision approach. Experiments are carried out with both grey scale and colour images of various resolutions, texture and structures. The proposed scheme has high peak-signal-to-noise ratio and better visual quality in comparison to the standard median filter, modified decision based unsymmetrical trimmed median filter and improved fast peer-group filter with a varying noise density from 10 to 90%.

Photos are becoming more spread with digital age. Cameras, smart phones and Internet provide large dataset of images available to a wide audience. Assessing memorability of these photos is becoming a challenging task. Besides, finding the best representative model for memorable images will enable memorability prediction. The authors develop a new approach-based rule of photography to evaluate image memorability. In fact, they use three groups of features: image basic features, layout features and image composition features. In addition, they introduce a diversified panel of classifiers based on some data mining techniques used for memorability analysis. They experiment their proposed approach and they compare its results to the state-of-the-art approaches dealing with image memorability. Their approach experiment's results prove that models used in their approach are encouraging predictors for image memorability.

The existing visibility distance estimation algorithms in foggy situations use the region growing method to extract the vertical position of inflection point of image intensity changing. These algorithms have lower inflection point location accuracy for an image with a non-homogeneous road surface. To deal with these problems, this study presents a novel visibility distance measuring technique under foggy weather conditions. This method combines two major models: inflection point estimation (IPE) model and transmission refining (TR) model. The proposed IPE model based on transmission computation model derives a very useful relation between the transmission value of inflection points and the constant . In order to acquire the more accurate transmission map and vertical position of each inflection point, this study establishes an effective TR model. This model exploits the edge information of input images, in order to significantly reduce the effects of artefact. The proposed algorithm provides more accurate visibility distance estimation of an image with a non-homogeneous road surface than the well-known algorithm through qualitative evaluations in experiments. The experimental results also show that the TR model has better outcomes than the guided filter approach through qualitative and quantitative evaluations.

Image width is an important factor for making the partially recovered data perceptually meaningful in image file carving. The authors conduct a comprehensive comparison of the performance of the representative methods for estimating the JPEG image width. Experimental results show that the best methods based on pixels are always better than the best methods based on quantised discrete cosine transform (DCT) coefficients. To keep the good performance of the pixel-based methods when the correct quantisation tables are unavailable, the authors replace the correct quantisation tables with the standard ones. Experimental results certify that such a replacement has only a little effect on the performance of the pixel-based methods, the best of which still outperform the best methods based on quantised DCT coefficients. The two results indicate that it may be enough to just focus on the pixel-based methods for future work. Finally, they propose a pixel-based method, which derives the candidate image widths from the most likely adjacent minimum coded unit (MCU) pairs in the vertical direction. The candidate width which appears most frequently is chosen as the estimated image width. Experimental results show that the proposed method usually has the best performance when most MCUs of an image are recovered.

Classification of benign and malignant pulmonary nodules can provide useful indicators for estimating the risk of lung cancer. In this study, an improved random forest (RF) algorithm is proposed for classification of benign and malignant pulmonary nodules in thoracic computed tomography images. First, an improved random walk algorithm is proposed to automatically segment pulmonary nodules. Then, intensity, geometric and texture features based on the grey-level co-occurrence matrix, rotation invariant uniform local binary pattern and Gabor filter methods are combined to generate an effective and discriminative feature vector. Mutual information is employed to reduce the dimensionality. Finally, an improved RF classifier is trained to classify benign and malignant nodules. An appropriate feature subset is selected by the bootstrap method and an effective combination method is introduced to predict a class label. The proposed classification method on the lung images dataset consortium dataset achieves a sensitivity of 0.92 and the area under the receiver-operating-characteristic curve of 0.95. An additional evaluation is performed on another dataset coming from General Hospital of Guangzhou Military Command. A mean sensitivity and a mean specificity of the proposed method are 0.85 and 0.82, respectively. Experimental results demonstrate that the proposed method achieves the satisfactory classification performance.

This study investigates local orientation-based approaches to the complex problem of pattern segmentation in three-dimensional (3D) texture image. The current problem focuses on the extraction of so-called lamellar colonies in titanium alloy, which, from the materials science and engineering point of view, are microstructural features that play a fundamental role on crack propagation and bifurcation during mechanical loading. Methods based on local orientation estimation extend the notion of using local gradient to reveal variation of semi-planar pattern orientation in the 3D image. The study introduces a computational approach that accelerates the calculation of the eigenvectors from the local matrices of inertia of all voxels composing the 3D image. Then different paths are proposed to segment colonies or inter-colony boundaries, i.e. polar orientation map and minimum scalar product map, in order to delimitate regions of similar orientations. The investigated segmentation methods have been compared with other methods that are mainly based on the popular solution of filter banks. Tests, which have been performed on both synthetic and real 3D images, show that the proposed local orientation-based methods better delineate object boundaries than the counterparts.

The main challenges of image encryption are robustness against attacks, key space, key sensitivity, and diffusion. To deal with these challenges, a differential evolution-based image encryption technique is proposed. In the proposed technique, two concepts are utilised to encrypt the images in an efficient manner. The first one is Arnold transform, which is utilised to permute the pixels position of an input image to generate a scrambled image. The second one is differential evolution, which is used to tune the parameters required by a beta chaotic map. Since the beta chaotic map suffers from parameter tuning issue. The entropy of an encrypted image is used as a fitness function. The proposed technique is compared with seven well-known image encryption techniques over five well-known images. The experimental results reveal that the proposed technique outperforms the other existing techniques in terms of security and better visual quality.

The increasing risks of border intrusions or attacks on sensitive facilities and the growing availability of surveillance cameras lead to extensive research efforts for robust detection of pedestrians using images. However, the surveillance of borders or sensitive facilities poses many challenges including the need to set up many cameras to cover the whole area of interest, the high bandwidth requirements for data streaming and the high-processing requirements. Driven by day and night capabilities of the thermal sensors and the distinguished thermal signature of humans, the authors propose a novel and robust method for the detection of pedestrians using thermal images. The method is composed of three steps: a detection which is based on a saliency map in conjunction with a contrast-enhancement technique, a shape description based on discrete Chebyshev moments and a classification step using a support vector machine classifier. The performance of the method is tested using two different thermal datasets and is compared with the conventional maximally stable extremal regions detector. The obtained results prove the robustness and the superiority of the proposed framework in terms of true and false positives rates and computational costs which make it suitable for low-performance processing platforms and real-time applications.