Image forgery detection using traditional algorithms takes much time to find forgeries. The new emerging methods for the detection of image forgery use a deep neural network algorithm. A hybrid deep learning (DL) and machine learning-based approach is used in this study for passive image forgery detection. A DL algorithm classifies images into the forged and not forged categories, whereas colour illumination localises forgery. The simulated results are compared to other algorithms on public datasets. The simulated results achieved 99% accuracy for CASIA1.0, 98% accuracy for CASIA2.0, 98% accuracy for BSDS300, 97% accuracy for DVMM, and 99% accuracy for CMFD image manipulation dataset.

Mammography is a very useful tool to diagnose breast cancer in early stages when it is easier to treat. There are two types of evidence that radiologists look for in a mammogram, calcifications and the existence of masses. In this study, an intelligent computer-aided diagnosis system is proposed for the detection of masses in mammographic images regardless of their nature. The proposed method uses a combination of extended maxima transformations, having different threshold values, in order to find suitable internal and external markers for a marker-based watershed segmentation. After segmentation, a two-stage classifier is used to distinguish the masses better from the healthy breast tissue. A feature vector based mainly on contrast and texture features is calculated and two alternative approaches, a Bayesian classifier and a support vector machine (SVM) with Gaussian kernel function, are implemented for further reduction of the false positive areas. The system was evaluated using the data from two online databases. Specifically, 73 mammographic images from the new curated breast imaging subset of digital database for screening mammography (CBIS-DDSM) database and all the mammographic images that contain masses from the mini-mammographic image analysis society (MIAS) database were used. The overall sensitivity, in both datasets, was near 80% when the Bayesian classifier was used and above 85% when the SVM was applied.

Brain tumours have increased rapidly in recent years as in other tumour types. Therefore, early and accurate diagnosis of brain tumour is vital for treatment. Magnetic resonance imaging (MRI) and histopathological assessments are the most common methods used in the detection of brain tumours. The research studies on non-invasive imaging methods such as MRI and magnetic resonance spectroscopy (MRS) have become widespread in recent years for brain tumour detection. In this study, a computer-assisted method is proposed for automatic grading of brain tumours on MRS signals. The classification of brain tumours with different grades is performed using long short term memory (LSTM) neural networks. In addition, additional features from MRS signals based on spectral entropy and instantaneous frequency are extracted. As a result of the experimental studies on the international MRS database (INTERPRET), it is seen that grading is achieved using the proposed method with average accuracy of 98.20%, sensitivity of 100%, and specificity of 97.53% performance results in three test studies carried out for the classification of brain tumour. Furthermore, in the grading of brain tumours using the proposed method, the average area under of the receiver operating characteristic curve is measured with high performance of 0.9936.

The study presents a novel approach, based on deep learning, for diagnosis of Parkinson's disease through medical imaging. The approach includes analysis and use of the knowledge extracted by deep convolutional and recurrent neural networks when trained with medical images, such as magnetic resonance images and dopamine transporters scans. Internal representations of the trained DNNs constitute the extracted knowledge which is used in a transfer learning and domain adaptation manner, so as to create a unified framework for prediction of Parkinson's across different medical environments. A large experimental study is presented illustrating the ability of the proposed approach to effectively predict Parkinson's, using different medical image sets from real environments.

As an analytic tool in medicine, particularly in radiology, deep learning is gaining much attention and opening a new way for disease diagnosis. Nonetheless, it is rather challenging to acquire large-scale detailed labelled datasets in the field of medical imaging. In fact, transfer learning provides a possible way to resolve this issue to a certain extent such that the parameter learning of a neural network starts with its pre-trained weights learned from a large-scale dataset of certain similar task, and fine-tunes on a small comprehensively annotated dataset for the particular target task. The main aim of this study is to apply the deep learning model to detect the synovial fluid of human knee joint from magnetic resonance images. A specialized convolutional neural network architecture is proposed for automated detection of human knee joint's synovial fluid. Two independent datasets are used in the training, development, and evaluation of the proposed model. It is demonstrated by the experimental results that the proposed model obtains high sensitivity, specificity, precision, and accuracy to the detection of human knee joint's synovial fluid. As a result, this proposed approach provides a novel and feasible way for automating and expediting the synovial fluid analysis.

Hyperspectral satellite imagery (HSI) is an advanced technology for object detection because it provides a large amount of information. Thus, the classification of HSIs is very complicated, so the methods of reducing spectral or spatial information generally degrade the quality of classification. In order to solve this problem and guarantee faster and more efficient processing, we propose a smart feature extraction (SFE) and classification by convolutional neural network (2D-CNN) method made up of two parts. The first consists in reducing spectral information by a probabilistic method based on the Softmax function. The second is classification by processing batches of data in the proposed CNN network. The method was tested on two public hyperspectral images (Indian Pines and SalinasA) to prove its effectiveness in increasing classification accuracy and reducing computing time.

This study investigates the use of binary features and fractal dimension analysis to evaluate the dispersion quality of nanofillers in thin polymeric films by using light microscopy images. For this purpose, polymeric films were cast with the inclusion of various montmorillonite (MMT) nanofiller amounts. Then the light microscopy images were captured from the polymeric films then preprocessed for the evaluation. Thresholding process was applied to the obtained images for each nanofiller percentage level. The obtained binary level images were used in the feature extraction process with binary statistics and fractal dimension. Thermogravimetric analysis (TGA) was used to evaluate the flame resist behaviour of polymeric films based on the dispersion quality of nanofillers. The samples with various nanofiller contents were tested using the image processing method and the results were all compared with the TGA results. The results obtained by the feature extraction process and TGA, about the dispersion quality of nanofillers, were all in good agreement.

To better restore a clean image from a noise observation under high noise levels, the authors propose an image denoising network based on the combination of multi-scale and residual learning. Instead of using filters with different large sizes in traditional multi-scale schemes, they arrange multi-layer convolutions with the filters of the same size to speed up the model. Some dilated convolutions of different rates are combined with the common convolutions to enrich the extracted features in multi-layer convolutions. Furthermore, they cascade the multi-layer convolutions with residual blocks to improve the performance of image denoising. Their extensive evaluations on several challenging datasets demonstrate that the proposed model outperforms the state-of-art methods under all different noise levels in terms of peak signal-to-noise ratio, and the visual effects achieved by the proposed model are also better than the competing methods.

Degraded Subterranean environments are an attractive case for miniature aerial vehicles, since there is a constant need to increase the safety operations in underground mines. The starting point for integrating aerial vehicles in the mining process is the capability to reliably navigate along tunnels. Inspired by recent advancements, this paper presents a collection of different, experimentally verified, methods tackling the problem of MAVs heading regulation while navigating in dark and textureless tunnel areas. More specifically, four different methods are presented in this work with the common goal to identify open space in the tunnel and align the MAV heading using either visual sensor in methods a) single image depth estimation, b) darkness contour detection, c) Convolutional Neural Network (CNN) regression and 2D Lidar sensor in method d) range geometry. For the works a)-c) the dark scene in the middle of the tunnel is considered as open space and is processed and converted to yaw rate command, while d) examines the geometry of the range measurements to calculate the yaw rate command. Experimental results from real underground tunnel demonstrate the performance of the methods in the field, while setting the ground for further developments in the aerial robotics community.

Fish-cage dysfunction in aquaculture installations can trigger significant negative consequences affecting the operational costs. Low oxygen levels, due to excessive fooling's, leads to decrease growth performance, and feed efficiency. Therefore, frequent periodic inspection of fish-cage nets is required, but this task can become quite expensive with the traditional means of employing professional divers that perform visual inspections at regular time intervals. The modern trend in aquaculture is to take advantage of IT technologies with the use of a small-sized, low-cost autonomous underwater vehicle, permanently residing within a fish cage and performing regular video inspection of the infrastructure for the entire net surface. In this study, we explore specialised image processing schemes to detect net holes of multiple area size and shape. These techniques are designed with the vision to provide robust solutions that take advantage of either global or local image structures to provide the efficient inspection of multiple net holes.

Three-dimensional (3D) local feature detection and description techniques are widely used for object registration and recognition applications. Although several evaluations of 3D local feature detection and description methods have already been published, these are constrained in a single dimensional scheme, i.e. either 3D or 2D methods that are applied onto multiple projections of the 3D data. However, cross-dimensional (mixed 2D and 3D) feature detection and description are yet to be investigated. Here, the authors evaluated the performance of both single and cross-dimensional feature detection and description methods on several 3D data sets and demonstrated the superiority of cross-dimensional over single-dimensional schemes.

Digital watermarking embeds a watermark to minimise the problem of illegal copying and disseminating multimedia contents. However, the existing techniques do not maintain the imperceptibility and robustness simultaneously. To achieve the same, this study proposes an optimised robust watermarking technique using chaotic kbest gravitational search algorithm. The chaotic kbest gravitational search algorithm is used to obtain the optimal values of embedding factors. The efficacy of the proposed technique has been experimented on a standard images and compared with the six recent state-of-the-art techniques in terms of imperceptibility and robustness. The experimental results validate that the proposed technique outperforms the other considered techniques.

The data hiding method embeds the data into covers such as video, audio, and images, which are used for integrity authentication, media protection, communication covert, copyright protection etc. In this work, several reversible data hiding (RDH) algorithms are analysed. Here, RDH algorithms are classified into six categories. They are histogram shifting centred RDH, code division multiplexing-based RDH, compression-based RDH, contrast enhancement with RDH, and expansion-based RDH, and RDH for encrypted images. In RDH,no information is lost when the message is recovered as the message is embedded into the host image. The embedded message is extracted using an extraction procedure. This study critically reviews the technological advancement for the past decade on the aspects of classification, pre-processing, feature extraction, and analysis technique. The current trend in RDH is to insert confidential data into a video by using RDH in the encoded area that has a wide selection of applications on the confidential communication inside the area of cybersecurity.

To improve the performance of the high-voltage copper contact burr image segmentation, a block cosparsity overcomplete learning transform image segmentation algorithm based on burr model is proposed in this study. In this study, k-means clustering method is used to initialise the clustering results; the authors found the algorithm is very effective for burr image processing in production process and the sparse overcomplete transform matrix is initialised by discrete cosine transform. The algorithm is expressed by a set of transforms. When the set of transforms is fixed, the penalty is corresponding to the condition number. A new burr model is proposed in this study. The parameters of the burr are the factors on infection of the sparse-level constant and the regularisation coefficient of the block cosparsity overcomplete learning transform algorithm. The algorithm divides all pixels into several groups. To evaluate the performance of the model, a large number of experiments have been carried out, and three image segmentation evaluation criterions have been used to evaluate the effectiveness of the algorithm. Experimental results show that this method is excellent in retaining weak edge information and avoiding the influence of three-dimensional structure compared with other algorithms.

A novel image multi-encryption architecture based on hybrid keystream sequence generated by a single hyperchaotic system and Haar discrete wavelet transform (HDWT) is proposed. The architecture consists of a pre-cipher stage, first encryption operation, Haar discrete wavelet decomposition stage and a second encryption operation. In the pre-cipher stage, the algorithm applies two-level pixel position permutation of the image. The first encryption operation is accomplished by diffusing the pixel's numbers with keystream sequence generated by a sequence generator SG-1. The resulting cipher image is decomposed using two-dimensional HDWT decomposition technique. The decomposed image is further encrypted through operation between keystream sequence generated by sequence generator SG-2 and bytes of the decomposed image which are selected with the aid of a novel pseudo – 4 bit Boolean truth table-based byte selection mechanism. SG-1 and SG-2 are special cases of the hyperchaotic system and are evolved online by separately nullifying selected parameters of the hyperchaotic system. The novelty of the proposed architecture lies in the possibility of increasing the number of sequence generators, which can result in exponentially huge key space and fast encryption speed. A comprehensive complexity analysis of performance, security and robustness to attacks, confirmed the feasibility of the architecture.

Image forgery poses a serious threat in electric power, medicine and other fields. Relevant departments need to pay a great price to identify the authenticity of the image. For traditional copy-move forgery image detection, the existing methods have at least two problems: low robustness and poor matching caused by a low number of feature points. Here, a novel similarity metric combining cosine and Jaccard is proposed to improve feature matching, which combines with oriented features from accelerated segment test and rotated binary robust independent elementary features (ORB) feature extraction to realise effective and fast image forgery detection. First, the image is divided into overlapping blocks, and ORB is used to extract the feature points of each image block to obtain the text information. Second, the novel similarity metric is used to calculate similarity and match the text. Finally, two image blocks with the highest similarity are located. The experimental results show that, on the one hand, ORB can greatly lessen detection time. On the other hand, the novel similarity metric can improve the poor matching caused by the small number of feature points. Combining the two methods can exhibit high robustness to translation, rotation, noise, illumination and JPEG compression.

Sign language recognition becomes a popular research field in human–computer interaction. Attention on hand signal analysis helps to make easy communication among computer and human for information sharing. Major focus of the gesture recognition system is to identify and recognise various gestures, by a computer. This study introduces optimisation of both classifier and feature set for static American sign language recognition. Initially, the hand part is segmented from other parts of the image through effective edge and skin colour detection. Thereafter, robust features are obtained using discrete cosine transform, Zernike moment, scale-invariant feature transform, speeded-up robust features, histogram of oriented gradients and binary object features from the segmented hand image. From these extracted features, an optimal feature set is selected by social ski driver optimisation algorithm. Deep Elman recurrent neural network classifier is then introduced for recognition purpose. Optimisation is performed on feature sets, derived by fusion of features obtained from the above methods, based on precision, accuracy, F-measure and recall. Finally, optimised feature set and best classifier are used to recognise the hand gesture for classification purpose. The performance of this proposed method is evaluated and compared with existing literature.

The authors describe and evaluate a method that detects ridges (symmetric axes) in an Euclidean distance map. The method detects ridge-pixels with a local-maxima search using only relational operations and has therefore minimal complexity. The resulting ridges exhibit a height profile that is suitable for region abstraction by means of simple parameterisation. The method is firstly evaluated on artificial stimuli with systematic shape variations using spatial jitter and contour fragmentation. Then it is mentioned, how the method can be used for developing region abstractions. Such abstractions have been already exploited in two tasks, hand-written digit identification (MNIST database) and image classification (satellite images and images of urban/natural landscapes); the classification results of those systems are competitive.

The main objective of this study is to compare face recognition accuracies in the case when the grey levels in each pixel of the face images are replaced by the gradient and the surface normal vectors. Extensive information is provided to explain the differences between the gradient and the proposed features. Some well-known face recognition methods, such as common vector approach (CVA), discriminative CVA, and support vector machines are applied to the well-known databases of AR and Yale for comparison other than introducing a new method what the authors called as Sum of Pixel Slope Similarities Approach. The authors’ experimental results are compared with the state-of-the-art methods to the best of their knowledge. In conclusion, their results imply that using the surface normal vectors rather than the gradient vectors in each pixel with no additional work on their elements gives better recognition rates.

In recent years, dictionary learning has shown to be an efficient tool in recovering images from their degraded, damaged or incomplete version. Especially, for medical images that contain significant details and characteristics. In this work, the authors are interested in this unsupervised learning technique for discovering and visualising the underlying structure of a medical image. Therefore, an adaptive bi-dictionary learning model for recovering magnetic resonance (MR) image from undersampled measurements is introduced. The proposed model learns two dictionaries, one over the underlying image and the other over its sparse gradient. Hence, the algorithm minimises a linear combination of three terms corresponding to the least-squares data fitting, dictionary learning over the pixel domain, and gradient-based dictionary. Numerically, experimental results on several MR images demonstrate that the proposed bi-dictionary framework can improve reconstruction accuracy over other methods.

With the rapid increase in the cases of deadly skin cancer, the classification on different types of skin cancer has been emerging as one of the most significant issues in the field of medical image. Several approaches have been proposed to help in diagnosing the categories of the skin lesions by means of traditional features or leveraging the widely used deep learning models. However, there are lack of the integrated frameworks to combine the hand-crafted traditional features and the deep Conv-features. Furthermore, the effective way to extract global and local features is also conducive to distinguish the specific lesions from normal skin. Hence, in this study, the authors present an integrated model to acquire more representative global–local features including the traditional local binary pattern features and deep Conv-features. In addition, several fusion strategies have conducted on the Global-DNN and Local-DNN for better performance. In order to extract more explicit features from the specific lesion areas, a target segmentation method based on visual saliency detection is employed to eliminate the background interference. Experimental results on ISIC-2017 skin cancer dataset demonstrate that the proposed Global-DNN and Global-Local models can obtain more effective feature representation which achieve outperformed results for skin cancer classification.

Sonar images are valuable in exploring underwater environmental information. As these images are generally limited by the viewing angle, sonar image mosaicking becomes an important research topic. By combining several frames consecutively acquired while the underwater vehicle is manoeuvring, an image with a wider view can be obtained. This study presents a fast sonar image mosaicking approach consisting of denoising, feature extraction, initial matching, splicing, and optimisation. Based on the Euclidean distance between initially matched points and dip angle of connection line, poorly matched feature point pairs are removed to avoid false matching. This way, the success rate of image mosaicking and quality of the resulting mosaicking is effectively improved.

This study proposes a framework of video coding based on Laplacian eigenmaps (LEM) and its related embedding and reconstruction algorithm (ERA). Firstly, a one-dimensional (1D) representation of LEM is adopted to achieve an extremely low bit per pixel (BPP). Secondly, dual k-nearest neighbours, which keeps neighbour relationships both in high-dimensional data space and low-dimensional representation space and overcomes the disadvantage of classical non-linear dimensionality reduction methods which cannot preserve the neighbour properties in both of the spaces, based ERA of LEM is employed to gain extraordinarily high peak-signal-to-noise ratio (PSNR). Thirdly, a unified framework of video coding is fit for intra-frame, inter-frame and multi-view video coding. Finally, it is evaluated by simulation experiments that, in the situation of low bitrate transmission, the proposed method can attain better performance of BPP and PSNR than that of the state-of-the-art methods, such as highly efficient video coding.

In the standard bag-of-visual-words model, the relationship between visual words and geometric structure information embedding in Voronoi cells is important for expressing the topology of the feature space. However, this information is usually ignored by recent works. To overcome it, the authors proposed a hybrid heterogeneous structure model (HHSM), where local hyperspheres and local structure subspaces are applied to simulate the intrinsic structure of the feature space. Firstly, the local hypersphere is formed by choosing some links between parts of visual words, with the use of a proposed decision strategy derived from k-dense neighbour algorithm. In order to capture the geometric structure information around the visual word, they then construct the local structure subspace with the transformed PCA principal vectors of the visual features within a Voronoi cell. Finally, this study introduces a novel feature encoding method based on the HHSM. Experiments are conducted on 15-Scenes, Pascal VOC2007, Caltech101, Caltech256 and MIT Indoor 67 datasets, which include 4485, 9963, 9146, 30607 and 15620 images, respectively. The results demonstrate the effectiveness of the proposed method in improving the accuracy of the classification. In addition, the proposed method achieves comparable performance when combined with CNN local features.

Exploitation of both spectral and spatial information in hyperspectral imagery is important for effective classification. Considering the cubical arrangement of data, the three-dimensional (3D) techniques could be effectively used to model hypespectral features. In this study, the 3D geometric moments are used to extract the rotation, scale, and translation invariant features. Unlike 2D moments, the 3D moments characterise the joint spectral–spatial properties. A classification method is proposed that uses the features derived from 3D geometric moments without vectorising or changing the original structure of the raw hyperspectral image. Unlike many other methods, the new method does not need a separate step for spectral feature extraction or dimensionality reduction. The moments are computed on the raw image that generate a comprehensive and smaller feature set. The experimental results from five benchmark airborne hyperspectral images demonstrate that the 3D moment based method yields good classification results better or comparable to several state-of-the-art methods.

Image super resolution refers to a class of signal processing algorithms to post-process a captured image to obtain its high resolution version. Multi-frame super resolution synthesises high resolution image from multiple low resolution observations. Performance of super resolution algorithms are adversely affected by the noise present in the input images. To develop a noise robust multi-frame image super resolution, an objective function is formulated which contains a weighted data fidelity term and a regularisation term consisting of a bilateral total variation (BTV) term and structure tensor total variation (STV) term. Both BTV and STV are weighted appropriately in a per pixel basis in such a way that the BTV contributes more in smooth regions and STV contributes more on the edges. These terms ensure the continuity of edges and the smoothness of flat regions. An adaptive weighting scheme with the data fidelity term helps to select the reliable pixel alone in the reconstruction process. The proposed method is experimentally evaluated for its performance in real data and different types of noises.

In this study, the authors investigate content based image retrieval (CBIR) using ordered-dither block truncation coding (ODBTC) and phase congruency feature (PCF). Relevant feature extraction plays a vital role for retrieving the image in CBIR. The unique reason to choose PCF with ODBTC is that it detects the edges and corners during variation of image while preserving image brightness and contrast. Combining the PCF and ODBTC features improves CBIR system usage in various visual data processing domains. Thus, yields a better CBIR system which assists in the reduction of storage space, decreases retrieval time and increases accuracy of the system. The precision and recall are used as performance metrics to evaluate the proposed method based on retrieval of relevant images. Extensive experimental results with Corel 1 K (1000 images), Corel 10 K (10000 images) and CALTECH 256 (30144 images) proves that the proposed method is more desirable than antecedent proposed CBIR system in terms of accuracy, precision and recall.

Nowadays, lung cancer is the leading cause of cancer death in both men and women. The early detection of potentially cancerous cells is the best way to improve the patient's chances of survival. In the medical field, computed tomography (CT) is the best imaging technique and it is helpful for doctors to accurately find the cancerous cells. The authors propose an automatic approach to analyse and segment the lungs and classify each lung into normal or cancer. Initially, the CT lung image is pre-processed to remove noise. Then, they combine the histogram analysis with thresholding and morphological operations to segment and extract the lung regions. In feature extraction stage, the radiomic features of each lung image are extracted separately. Then to improve the classification accuracy, some of the optimum features are selected using modified graph clustering-based whale optimisation algorithm. Finally, the selected features are classified using ensemble classifiers such as support vector machine, K-nearest neighbour, and random forest. Experimental result demonstrates that the proposed method achieves better performance in terms of sensitivity, specificity, precision, recall, F-measure, and accuracy when compared with other state-of-art approaches.

In order to realise autonomous navigation of unmanned platforms in urban or off-road environments, it is crucial to study accurate, versatile and real-time road detection methods. This study proposes an adaptive road detection method that combines lane lines and obstacle boundaries, which can be applied to a variety of driving environments. Combining multi-channel threshold processing, it is robust to lane feature detection under various complex situations. Obstacle information extracted from the grid image constructed by 3D LIDAR point cloud is used for lane feature selection to avoid interference from pedestrians and vehicles. The proposed method makes use of adaptive sliding window for feature selection, and piecewise least squares method for road line fitting. Experimental results on dataset and in real-world environments show that the proposed method can overcome illumination changes, shadow occlusion, pedestrian, vehicle interference and so on in a variety of scenes. The proposed method has good enough efficiency, robustness and real-time performance.

The discriminative correlation filter (DCF) method is widely used in target tracking due to its real-time performance. However, the computational efficiency of DCF results in boundary effect, which reduces the tracking accuracy in fast motion scene. Besides, background noise is always required to be carefully handled for they will cause trouble in scenes such as background clutter, occlusion, deformation etc. To address the two issues, this study proposes masked discriminative correlation filter, which uses mask to process DCF filter as well as target samples so as to suppress boundary effect and background noise. Experimental results on benchmark datasets show that the proposed tracker performs better than a series of benchmark trackers, and is superior to them in almost various scenes.

In this study, the authors introduce a novel deep generative adversarial networks (GANs) model for global mosaic removal. The methods used in the proposed study consist of GANs model and a novel algorithm for maintaining and repairing (MR) images. The conventional mosaic removal algorithms all employ the correlation between the inserted pixel and its neighbouring pixels, which have a limited effect on the local mosaic removal but do not work well for the global mosaic removal. To respond to this difficulty, the authors introduce an MRGAN model with two novel parsing networks. Unlike previous GANs, the MR algorithm is used to calculate the pixel loss and content loss. The experimental comparison results show that the proposed MRGAN model has achieved leading results for the global mosaic removal task.

The most important part of digital image analysis is object classification. Nowadays, deep learning makes an enormous achievement in computer vision problems. So there has been a lot of interests in applying features learned by convolutional neural networks (CNNs) on general image recognition to more tasks such as object detection, segmentation and face recognition. Leukocoria detection is one of the serious challenges in infant retinal treatment. Leukocoria is represented as an abnormal white reflection appearing in the eyes of an infant suffering from retinoblastoma. This research proposes a deep Visual Geometry Group-net CNN classifier for automatic detection of leukocoria. The proposed classifier comprises pre-processing, feature extraction and classification. The deep CNN classifier contains convolution layer, pooling layer and fully connected layer with weights are developed on each image. Experimental results based on several eye images consist of ordinary and leukocoric from flicker, and it demonstrates that the proposed classifier provides better results with the accuracy of 98.5% and the error rate is below 2% which exceeds the current results.

Salient object detection is evaluated using binary ground truth (GT) with the labels being salient object class and background. In this study, the authors corroborate based on three subjective experiments on a novel image dataset that objects in natural images are inherently perceived to have varying levels of importance. The authors' dataset, named SalMoN (saliency in multi-object natural images), has 588 images containing multiple objects. The subjective experiments performed record spontaneous attention and perception through eye fixation duration, point clicking and rectangle drawing. As object saliency in a multi-object image is inherently multi-level, they propose that salient object detection must be evaluated for the capability to detect all multi-level salient objects apart from the salient object class detection capability. For this purpose, they generate multi-level maps as GT corresponding to all the dataset images using the results of the subjective experiments, with the labels being multi-level salient objects and background. They then propose the use of mean absolute error, Kendall's rank correlation and average area under precision–recall curve to evaluate existing salient object detection methods on their multi-level saliency GT dataset. Approaches that represent saliency detection on images as local-global hierarchical processing of a graph perform well in their dataset.

This study proposes an automatic recognition and location system of electric vehicle charging port with application to automatic charging. The system obtains the charging port posture through image processing, and performs the insertion motion in combination with the robot arm to complete the charging gun insertion of the automatic charging link. The framework of the system is mainly divided into three parts, recognition, location and insertion. In the charging port recognition, the convolutional neural network-based method is used, and the recognition success rate is up to 98.9% under the light intensity of 4000 lux; in the location of the charging port, the method of solving the pose based on the circle feature is adopted. The average value of the position error is within 1.4 mm, and the average value of the attitude angle error is within 1.6°, which meets the accuracy requirement of the insertion experiment; in the charging gun insertion, the motion of the robot is planned by interpolation algorithm. The lower limit of the successful insertion is about 135 lux and the upper limit is about 9350 lux.