Complex Background Images Research Articles

An Image-Based Uninterrupted Lagrangian Tracking Method for Measurement of Sediment Motion at Grain Scale

The imaging technique provides an efficient non-intrusive way for studying local sediment transport with a low rate in open-channel flows. It aims to track all sediment trajectories above the background consists of similar particles (i.e., top-view images of the channel). For this area of interest, currently used imaging methods can be summarized as a two-step framework for identifying and matching active bed-load particles. While effective against a simple and clear background, the two-step approach fails to yield accurate and uninterrupted Lagrangian paths for the sediment particles against complex image background consists of similar particles. This study presents a three-step approach to improve the accuracy of the tracking method. The first two steps of this approach based on image subtraction, centroid exaction and Kalman filter entail improvements to those of the classic methods. The third step based on the nearest neighbor algorithm and spline interpolation manage to identify broken chains and connect them to reconstruct uninterrupted Lagrangian paths of the sediment particles. The verification against simulated images and experimental data shows that the improved three-step approach yields more accurate estimation of bed-load kinematics than the classic two-step method.

A new Connected Component Analysis based System for Text Segmentation in Degraded Historical Document Images

Historical documents contain valuable heritage information. These documents are preserved in the manuscript preservation center and archaeological departments. They are mostly degraded in nature and hence hard to read and understand the contents. So, there is a need for text segmentation and feature extraction to convert these manuscripts into machine editable format. In this work, we present an effective way to segment historical document images into characters. It is a challenging segmentation process due to complex background images. In this paper, horizontal histogram, vertical histogram and connected component analysis is used to segment text documents images. In this algorithm, the input image is converted to gray scale image, then gray image is converted into binary image [Otsu’s method] and then all the objects containing fewer than desired pixels are removed. Line and word segmentation is implemented using horizontal and vertical histogram method respectively. Then the connected components are labeled and properties are measured for the image regions. Connected component analysis is used to segment the characters and the individual characters are extracted. The simulation result shows that the proposed segmentation method achieves an average accuracy of 93.37% for HDLAC 2011 DATASET. Moreover this method is more efficient and more suitable for real time tasks.

Crop leaf disease recognition based on Self-Attention convolutional neural network

The characteristics of the complex background in crop disease image, the small disease area, and the small contrast between the disease region and the background that easily causes confusion between them, seriously affect the recognition robustness and accuracy. To address these issues, we propose a Self-Attention Convolutional Neural Network (SACNN), which extracts effective features of crop disease spots to identify crop diseases. Our SACNN includes a basic network and a self-attention network: the basic network is for extracting the global features of the image, and the self-attention network is for obtaining the local features of the lesion area. Extensive experimental results show that the recognition accuracy of SACNN on AES-CD9214 and MK-D2 is 95.33% and 98.0%, respectively. The recognition accuracy of SACNN on MK-D2 has outperformed the state-of-the-art method by 2.9%, which implies that the CNN with self-attention can focus on the important areas of the image, and thus can improve the recognition accuracy. Adding different levels of noise to the AES-CD9214 test set shows the anti-interference ability and the strong robustness of SACNN. In addition, we discuss the influence of the location selection, channel size setting, network number and other aspects of the self-attention network on the recognition performance, in order to show the self-attention network working mechanism and provide inspiration for future research.

Complex-background Image De-noising Algorithm Based on Wavelet Transform

Complex-background Image De-noising Algorithm Based on Wavelet Transform

Color disease spot image segmentation algorithm based on chaotic particle swarm optimization and FCM

Aiming at solving the problems of complex image background and difficulties in the later image segmentation, an image segmentation algorithm based on the chaotic particle swarm algorithm and fuzzy clustering is proposed. First, the color space is converted from the RGB color space into the HIS color space. Then, a hybrid algorithm consisted of chaotic particle swarm optimization and fuzzy clustering is introduced. Each color component is processed by the algorithm, and the corresponding partition graph is obtained. Finally, the color space is converted into the RGB color space to achieve the segmentation effect. Experimental results show that the new algorithm has higher accuracy to segment the image and has good robustness to noises.

Martial Arts Routine Difficulty Action Technology VR Image Target Real-Time Extraction Simulation

With the gradual increase in the difficulty of competitive martial arts, athletes must complete fine, stable, high-quality and difficult movements in order to achieve excellent performance. The real-time extraction of martial arts movements is a topic that many martial arts enthusiasts care about. This study mainly discusses the real-time extraction and simulation of VR image target in the difficult movement technology of martial arts routine. Considering the complex characteristics of martial arts movements, this article will analyze the preprocessing content of existing images. This includes image enhancement and image filtering, and uses median filtering methods to enhance the characteristics of the collected images. In this way, the visual effect of the original image can be improved, and the processed image will contribute to the subsequent segmentation. A new image segmentation method is proposed for the color model of the image. According to the H component of the HSV model representing the characteristics of chromaticity, the color image is transformed into the HSV model, and the H component is extracted. The histogram concept applies to H components. Based on the histogram of the H component, the segmentation threshold is determined, and the cropping target in the image is detected. Because the model space is very sensitive to color, VR technology is used to automatically determine the segmentation target. Combined with the above division methods, the automatic extraction of objects in the image is completed. The method of using VR technology for image extraction processing has high precision, and the error value is 3.92%<5%. The research results show that the method has good segmentation results, and is suitable for image segmentation under complex background and automatic image extraction under complex background.

Pixel Intensity-Based Contrast Algorithm (PICA) for Image Edges Extraction (IEE)

In this paper, images’ pixels are exploited to extract objects’ edges. This paper has proposed a Pixel Intensity based Contrast Algorithm (PICA) for Image Edges Extraction (IEE). This paper highlights three contributions. Firstly, IEE process is fast and PICA has less computation time when processing different images’ sizes. Secondly, IEE is simple and uses a $2\times 4$ mask which is different from other masks where it doesn’t require while -loop(s) during processing images. Instead, it has adopted an if -conditional procedure to reduce the code complexity and enhance computation time. That is, the reason why this design is faster than other designs and how it contributes to IEE will be explained. Thirdly, design and codes of IEE and its mask are available, made an open source, and in-detail presented and supported by an interactive file; it is simulated in a video motion design. One of the PICA’s features and contributions is that PICA has adopted to use less while-loop(s) than traditional methods and that has contributed to the computation time and code complexity. Experiments have tested 526 samples with different images’ conditions e.g., inclined, blurry, and complex-background images to evaluate PICA’s performance in terms of computation time, enhancement rate for processing a single image, accuracy, and code complexity. By comparing PICA to other research works, PICA consumes 5.7 mS to process a single image which is faster and has less code complexity by $u\times u$ . Results have shown that PICA can accurately detect edges under different images’ conditions. Results have shown that PICA has enhanced computation time rate for processing a single image by 92.1% compared to other works. PICA has confirmed it is accurate and robust under different images’ conditions. PICA can be used with several types of images e.g., medical images and useful for real-time applications.

Assessment of Object Segmentation Techniques for Object Based Image Retrieval

Objects relates more to human perception than any other attributes of an image. Image segmentation is a significant image processing technique to get the objects from complex image background. This work assesses the techniques of segmentation from basic global thresholding, edge based methods up to the advanced techniques such as K-means, Active Contour Model (Snakes) segmentation approaches. Later, results are post processed with the help of morphological operations and make them suitable for object based image retrieval. It also provides the comparative analysis and empirical assessment of performance of the proposed modified segmentation approaches.

IR saliency detection via a GCF-SB visual attention framework

Infrared (IR) saliency detection with high detection accuracy is a challenging task due to the complex background and low contrast of IR images. In this paper, an IR saliency detection method via a new visual attention framework is proposed, which comprises two phases. In the first phase, a Gray & Contrast Features (GCF) model is established, in which the IR image is processed in two feature channels, a gray feature channel and a contrast feature channel. And then a primary feature map can be obtained by fusing the gray and contrast features from these two channels, which is the basis of the second phase. In the second phase, a Similarity-based Bayes (SB) model is established, in which two prior probabilities and two likelihood functions are calculated according to the previously obtained primary feature map. Finally, the saliency map is calculated with the obtained prior probabilities and likelihood functions by Bayes formula. Experimental results indicate that the proposed method can effectively reduce noise and enhance contrast of IR images with complex background and low contrast, and obtain a higher detection accuracy and robustness than seven state-of-the-art methods.

Active contour model based on LIF model and optimal DoG operator energy for image segmentation

In order to solve the problem that the region-based active contour model is sensitive to the initial contour position, the convergence is poor, and the active contour model can not obtain good segmentation results when segmenting complex background images and severe intensity inhomogeneous images. In this paper, an active contour model which combines Local Image Fitting (LIF) and Difference of Gaussian (DoG) operator energy for image segmentation is proposed. Firstly, an optimal DoG operator is obtained by using the edge energy term, it can enhance the edge while smoothing inhomogeneous regions. Then, using the DoG energy term which is obtained in the first step and the LIF energy term to construct the total function energy terms. In this process, the regularization term is also established, it can control the smoothness of evolution curve, avoiding over-segmentation and re-initialization step. Finally, the variational method and gradient descent flow method are adopted to minimize the total energy functional for segmentation. Compared with other region-based active contour models, the experimental results show that the proposed method can achieve a better segmentation performance with less iterations and high calculation efficiency while segmenting the synthetic and real images with intensity inhomogeneity.

YOLOv3-Lite: A Lightweight Crack Detection Network for Aircraft Structure Based on Depthwise Separable Convolutions

Due to the high proportion of aircraft faults caused by cracks in aircraft structures, crack inspection in aircraft structures has long played an important role in the aviation industry. The existing approaches, however, are time-consuming or have poor accuracy, given the complex background of aircraft structure images. In order to solve these problems, we propose the YOLOv3-Lite method, which combines depthwise separable convolution, feature pyramids, and YOLOv3. Depthwise separable convolution is employed to design the backbone network for reducing parameters and for extracting crack features effectively. Then, the feature pyramid joins together low-resolution, semantically strong features at a high-resolution for obtaining rich semantics. Finally, YOLOv3 is used for the bounding box regression. YOLOv3-Lite is a fast and accurate crack detection method, which can be used on aircraft structure such as fuselage or engine blades. The result shows that, with almost no loss of detection accuracy, the speed of YOLOv3-Lite is 50% more than that of YOLOv3. It can be concluded that YOLOv3-Lite can reach state-of-the-art performance.

Research on Obstacle Avoidance of Six-Rotor Aircraft Transmission Line Based on Image

Aiming at the problem that the real-time obstacle avoidance of power line by power line patrol UAV is difficult to achieve, this paper takes the image of power line captured by dynamic camera as the object, proposes an algorithm based on EDLines to identify the power line and realize the distance estimation with the power line according to the change of feature points. Firstly, the ED algorithm detects the edge of the image, and then a model is built according to the characteristics of the transmission line. The identified lines are screened and detected to reduce the interference of the complex background. At the same time, the linear growth method connects the discontinuous transmission lines. Finally, the feature points on the transmission line are extracted, the matching relationship between frames in the image sequence is used to track the feature points, and combined the position of UAV obtains the distance relationship, and the corresponding control strategy is proposed. Ordinary background and complex background images verifies the accuracy of the proposed algorithm, and the actual distance obtained is compared with the estimated distance. Experiments show that the matching accuracy and recognition speed of this algorithm are improved compared with Hough transform algorithm in the process of detecting transmission lines.

Fully automatic segmentation method for medicinal plant leaf images in complex background

Vein is a vascular bundle in a leaf which often extends from the leaf center to the edge. In this paper, an accurate and fully automatic segmentation method for medicinal plant leaf images in complex background is proposed by taking vein enhancement and extraction in the image as the core. It has laid a solid foundation for the non-destructive machine identification of medicinal plants taking a mobile phone as the terminal.Gradient magnitude image and gradient angle image are obtained by directly solving color RGB images. By fully applying the objective rule that the veins are linear and the consistency of gradient angles of adjacent vein points is relatively high, the veins in the gradient magnitude images are further enhanced by the standard deviation of the gradient angles to obtain the vein enhancement image. Based on this image, the OTSU method is used to obtain a binary image taking the veins as the foreground, and the main veins are detected from it. In other areas beyond the main veins in the vein enhancement map, fine veins are detected and then connected to the main veins. Then, a foreground marker image targeting the veins is obtained. The background marker image is segmented by using the OTSU method in each component image of the color image and then screened after a specific ratio is calculated. Based on the foreground markers and the background markers, the marker-controlled watershed method is applied to obtain the binary segmentation result.A series of experimental tests based on a self-built database and another widely used database show that the accuracy of the proposed method is better than many main fully automatic image segmentation methods including deep learning FCN. Its running speed is also fast and slightly lower than that of the OTSU threshold segmentation.

Detection of trace elements in metallurgical products under complex background based on visual saliency

ABSTRACT Aiming at the problem of detecting blurred targets by line scan CCD images in complex background, a target detection algorithm based on visual saliency is proposed. According to the characteristics of the line scan CCD target image, the feature of the weak target is analyzed, and the CCD image is preprocessed. By calculating the difference between two near frames in the image, all the complex backgrounds of static and slow motion are removed. A new feature vector is constructed, and a double search window similar to the target shape is designed. A significant image of the line scan CCD image is obtained by calculating the similarity of the feature vectors in the center of the window and the surrounding pixels. The salient map is then clustered and segmented to extract the target. The experimental results show that compared with weighted average filtering and median filtering, this method can effectively suppress noise and background, detect dark targets in images, increase the global signal-to-noise ratio by two times, and improve the detection reliability.

Single image dehazing using gradient channel prior

The dehazing techniques designed so far are not so-effective at preserving texture details, especially in case of a complex background and large haze gradient image. Therefore, the exploration of new alternatives for designing an effective prior is desirable. Thus, in this research work, Gradient profile prior (GPP) is designed to evaluate depth map from hazy images. The transmission map is also improved by utilizing Guided anisotropic diffusion and iterative learning based image filter (GADILF). The restoration model is also improved to reduce the effect of pixels saturation and color distortion from restored images. Performance analysis demonstrates that GPP can naturally restore the hazy image especially at the edges of sudden changes in the obtained depth map. Through extensive analysis, it has been found that GPP based dehazing can effectively suppress visual artefacts for hazy images and yield high-quality results as compared to the competitive dehazing techniques both quantitatively and qualitatively. Moreover, the relatively high computational speed of the proposed technique will facilitate it in real-time applications.

Pose detection of parallel robot based on improved RANSAC algorithm

For the factors of complex image background, unobvious end-effector characteristics and uneven illumination in the pose detection of parallel robot based on binocular vision, the detection speed, and accuracy cannot meet the requirement of the closed-loop control. So a pose detection method based on improved RANSAC algorithm is presented. First, considering that the image of parallel robot is rigid and has multiple corner points, the Harris–Scale Invariant Feature Transform algorithm is adopted to realize image prematching. The feature points are extracted by Harris and matched by Scale Invariant Feature Transform to realize good accuracy and real-time performance. Second, for the mismatching from prematching, an improved RANSAC algorithm is proposed to refine the prematching results. This improved algorithm can overcome the disadvantages of mismatching and time-consuming of the conventional RANSAC algorithm by selecting feature points in separated grids of the images and predetecting to validate provisional model. The improved RANSAC algorithm was applied to a self-developed novel 3-degrees of freedom parallel robot to verify the validity. The experiment results show that, compared with the conventional algorithm, the average matching time decreases by 63.45%, the average matching accuracy increases by 15.66%, the average deviations of pose detection in Y direction, Z direction, and roll angle [Formula: see text] decrease by 0.871 mm, 0.82 mm, and 0.704°, respectively, using improved algorithm to refine the prematching results. The real-time performance and accuracy of pose detection of parallel robot can be improved.

A Novel Dataset For Intelligent Indoor Object Detection Systems

Indoor Scene understanding and indoor objects detection is a complex high-level task for automated systems applied to natural environments. Indeed, such a task requires huge annotated indoor images to train and test intelligent computer vision applications. One of the challenging questions is to adopt and to enhance technologies to assist indoor navigation for visually impaired people (VIP) and thus improve their daily life quality. This paper presents a new labeled indoor object dataset elaborated with a goal of indoor object detection (useful for indoor localization and navigation tasks). This dataset consists of 8000 indoor images containing 16 different indoor landmark objects and classes. The originality of the annotations comes from two new facts taken into account: (1) the spatial relationships between objects present in the scene and (2) actions possible to apply to those objects (relationships between VIP and an object).This collected dataset presents many specifications and strengths as it presents various data under various lighting conditions and complex image background to ensure more robustness when training and testing objects detectors. The proposed dataset, ready for use, provides 16 vital indoor object classes in order to contribute for indoor assistance navigation for VIP.

Geospatial Object Detection on High Resolution Remote Sensing Imagery Based on Double Multi-Scale Feature Pyramid Network

Object detection on very-high-resolution (VHR) remote sensing imagery has attracted a lot of attention in the field of image automatic interpretation. Region-based convolutional neural networks (CNNs) have been vastly promoted in this domain, which first generate candidate regions and then accurately classify and locate the objects existing in these regions. However, the overlarge images, the complex image backgrounds and the uneven size and quantity distribution of training samples make the detection tasks more challenging, especially for small and dense objects. To solve these problems, an effective region-based VHR remote sensing imagery object detection framework named Double Multi-scale Feature Pyramid Network (DM-FPN) was proposed in this paper, which utilizes inherent multi-scale pyramidal features and combines the strong-semantic, low-resolution features and the weak-semantic, high-resolution features simultaneously. DM-FPN consists of a multi-scale region proposal network and a multi-scale object detection network, these two modules share convolutional layers and can be trained end-to-end. We proposed several multi-scale training strategies to increase the diversity of training data and overcome the size restrictions of the input images. We also proposed multi-scale inference and adaptive categorical non-maximum suppression (ACNMS) strategies to promote detection performance, especially for small and dense objects. Extensive experiments and comprehensive evaluations on large-scale DOTA dataset demonstrate the effectiveness of the proposed framework, which achieves mean average precision (mAP) value of 0.7927 on validation dataset and the best mAP value of 0.793 on testing dataset.

Saliency-Aware Convolution Neural Network for Ship Detection in Surveillance Video

Real-time detection of inshore ships plays an essential role in the efficient monitoring and management of maritime traffic and transportation for port management. Current ship detection methods which are mainly based on remote sensing images or radar images hardly meet real-time requirement due to the timeliness of image acquisition. In this paper, we propose to use visual images captured by an on-land surveillance camera network to achieve real-time detection. However, due to the complex background of visual images and the diversity of ship categories, the existing convolution neural network (CNN) based methods are either inaccurate or slow. To achieve high detection accuracy and real-time performance simultaneously, we propose a saliency-aware CNN framework for ship detection, comprising comprehensive ship discriminative features, such as deep feature, saliency map, and coastline prior. This model uses CNN to predict the category and the position of ships and uses the global contrast based salient region detection to correct the location. We also extract coastline information and respectively incorporate it into CNN and saliency detection to obtain more accurate ship locations. We implement our model on Darknet under CUDA 8.0 and CUDNN V5 and use a real-world visual image dataset for training and evaluation. The experimental results show that our model outperforms representative counterparts (Faster R-CNN, SSD, and YOLOv2) in terms of accuracy and speed.

RealPoint3D: An Efficient Generation Network for 3D Object Reconstruction From a Single Image

The generation of 3D models from a single image has recently received much attention, based on which point cloud generation methods have been developed. However, most current 3D reconstruction methods only work for relatively pure backgrounds, which limit their applications on real images. Meanwhile, more fine-grained details are required to provide finer models. This paper proposes an end-to-end efficient generation network, which is composed of an encoder, a 2D-3D fusion module, and a decoder. First, a single-object image and a nearest-shape retrieval from ShapeNet are fed into the network; then, the two encoders are integrated adaptively according to their information integrity, followed by the decoder to obtain fine-grained point clouds. The point cloud from the nearest shape effectively instructs the generation of finer point clouds. To have a consistent spatial distribution from multi-view observations, our algorithm adopts projection loss as an additional supervisor. The experiments on complex and pure background images show that our method attains state-of-the-art accuracy compared with volumetric and point set generation methods, particularly toward fine-grained details, and it works well for both complex backgrounds and multiple view angles.