Figure - available from: Multimedia Tools and Applications
This content is subject to copyright. Terms and conditions apply.
Source publication
In this paper, we will address the issue of detecting small target in a color image from the perspectives of both stability and saliency. First, we consider small target detection as a stable region extraction problem. Several stability criteria are applied to generate a stability map, which involves a set of locally stable regions derived from seq...
Similar publications
In recent years, deep learning has achieved re-markable success on a large number of tasks in machine learning. Despite this, many aspects of the functions learned by deep neural networks remain unknown. This study provides a general framework for analyzing and visualizing the operation of deep neural networks. We address these issues from an empir...
Objective:
To analyze the incidence of tuberculosis in indigenous and non-indigenous residents in the state of Pará from 2005-2013.
Method:
An ecological study was performed with data from SINAN, stratified for the 13 existing Regional Health Centers in Pará. The tuberculosis incidence rates were calculated for indigenous and non-indigenous popula...
This paper presents an approach for hybrid treemaps, which applies and combines several different layout principles within a single tree map in contrast to traditional treemap variants based on a single layout concept. To this end, we analyze shortcomings of state-of-the-art treemap algorithms such as Moore, Voronoi and Strip layouts. Based on a nu...
Citations
... The data holes in the TLS point cloud were divided into top voids and facade voids. Concerning the top surface hole, the study projected and rasterized the TLS point cloud from the perspective of a top-down view, taking the point cloud elevation value contained in the cell as the cell value, outputting the missing area of the top surface point cloud with zero cell value accordingly, and further employing the data segmentation method based on Euclidelist clustering [17] and alpha shape [18]. Other algorithms were utilized to screen the top surface point cloud voids and extract the boundary range of the void projection surface. ...
... In this research, a structure in the Lianhua Campus of Kunming University of Science and Technology in Kunming City, Yunnan Province was selected as the object, with The data holes in the TLS point cloud were divided into top voids and facade voids. Concerning the top surface hole, the study projected and rasterized the TLS point cloud from the perspective of a top-down view, taking the point cloud elevation value contained in the cell as the cell value, outputting the missing area of the top surface point cloud with zero cell value accordingly, and further employing the data segmentation method based on Euclidelist clustering [17] and alpha shape [18]. Other algorithms were utilized to screen the top surface point cloud voids and extract the boundary range of the void projection surface. ...
As one of the best means of obtaining the geometry information of special shaped structures, point cloud data acquisition can be achieved by laser scanning or photogrammetry. However, there are some differences in the quantity, quality, and information type of point clouds obtained by different methods when collecting point clouds of the same structure, due to differences in sensor mechanisms and collection paths. Thus, this study aimed to combine the complementary advantages of multi-source point cloud data and provide the high-quality basic data required for structure measurement and modeling. Specifically, low-altitude photogrammetry technologies such as hand-held laser scanners (HLS), terrestrial laser scanners (TLS), and unmanned aerial systems (UAS) were adopted to collect point cloud data of the same special-shaped structure in different paths. The advantages and disadvantages of different point cloud acquisition methods of special-shaped structures were analyzed from the perspective of the point cloud acquisition mechanism of different sensors, point cloud data integrity, and single-point geometric characteristics of the point cloud. Additionally, a point cloud void repair technology based on the TLS point cloud was proposed according to the analysis results. Under the premise of unifying the spatial position relationship of the three point clouds, the M3C2 distance algorithm was performed to extract the point clouds with significant spatial position differences in the same area of the structure from the three point clouds. Meanwhile, the single-point geometric feature differences of the multi-source point cloud in the area with the same neighborhood radius was calculated. With the kernel density distribution of the feature difference, the feature points filtered from the HLS point cloud and the TLS point cloud were fused to enrich the number of feature points in the TLS point cloud. In addition, the TLS point cloud voids were located by raster projection, and the point clouds within the void range were extracted, or the closest points were retrieved from the other two heterologous point clouds, to repair the top surface and façade voids of the TLS point cloud. Finally, high-quality basic point cloud data of the special-shaped structure were generated.
... Since there are very few brain-like location specific algorithms, we selected the brainlike saliency model constructed by Itti [39] based on human attention during visual search and the regional stability and saliency model proposed by Luo et al. [40] based on local center-surround difference and the global rarity of human visual perception for comparison. We consider the output saliency map of these two models as the location information. ...
Small object detection is one of the most challenging and non-negligible fields in computer vision. Inspired by the location–focus–identification process of the avian visual system, we present our location-focused small-object-detection algorithm for video or image sequence, OTNet. The model contains three modules corresponding to the forms of saliency, which drive the strongest response of OT to calculate the saliency map. The three modules are responsible for temporal–spatial feature extraction, spatial feature extraction and memory matching, respectively. We tested our model on the AU-AIR dataset and achieved up to 97.95% recall rate, 85.73% precision rate and 89.94 F1 score with a lower computational complexity. Our model is also able to work as a plugin module for other object detection models to improve their performance in bird-view images, especially for detecting smaller objects. We managed to improve the detection performance by up to 40.01%. The results show that our model performs well on the common metrics on detection, while simulating visual information processing for object localization of the avian brain.
... Considering the strong dependence relationship between the road and vehicles in the ground scene, the road areas in the image are extracted by segmentation networks to suppress the interference of complex backgrounds. Then tiny vehicles on the road are detected by the improved RSS [21] algorithm that incorporates stability region and saliency detection to strengthen the visual attention information. The motion information of vehicles is further employed to enhance the detection results with spatial-temporal confidence. ...
... In the process of changing the threshold, the region with the smallest area change is extracted as the maximally stable extremal region, to distinguish the object from the surrounding background. RSS [21] combines stability region detection and saliency detection to achieve small object detection in color images under simple background, but there are still problems of false alarms and missing objects. ...
... RSS [21] is a small object detection algorithm of color images that combines regional stability and saliency. It mainly includes three parts: stability region extraction, saliency detection, and integration of stability and saliency results. ...
Mid-to-high altitude Unmanned Aerial Vehicle (UAV) imagery can provide important remote sensing information between satellite and low altitude platforms, and vehicle detection in mid-to-high altitude UAV images plays a crucial role in land monitoring and disaster relief. However, the high background complexity of images and limited pixels of objects challenge the performance of tiny vehicle detection. Traditional methods suffer from poor adaptation ability to complex backgrounds, while deep neural networks (DNNs) have inherent defects in feature extraction of tiny objects with finite pixels. To address the issue above, this paper puts forward a vehicle detection method combining the DNNs-based and traditional methods for mid-to-high altitude UAV images. We first employ the deep segmentation network to exploit the co-occurrence of the road and vehicles, then detect tiny vehicles based on visual attention mechanism with spatial-temporal constraint information. Experimental results show that the proposed method achieves effective detection of tiny vehicles in complex backgrounds. In addition, ablation experiments are performed to inspect the effectiveness of each component, and comparative experiments on tinier objects are carried out to prove the superior generalization performance of our method in detecting vehicles with a limited size of 5 × 5 pixels or less.
... However, consideration of local relevance among neighboring regions can lead to incorrect suppression of salient regions, especially in images with heterogeneous salient object features [76]. A local contrast-based method for detecting small targets by computing contrast between the targeted small regions and surrounding regions was proposed [77]. Due to the limited spatial neighborhood consideration in the local contrast method, large salient regions can be easily excluded [42]. ...
Salient object detection represents a novel preprocessing stage of many practical image applications in the discipline of computer vision. Saliency detection is generally a complex process to copycat the human vision system in the processing of color images. It is a convoluted process because of the existence of countless properties inherent in color images that can hamper performance. Due to diversified color image properties, a method that is appropriate for one category of images may not necessarily be suitable for others. The selection of image abstraction is a decisive preprocessing step in saliency computation and region-based image abstraction has become popular because of its computational efficiency and robustness. However, the performances of the existing region-based salient object detection methods are extremely hooked on the selection of an optimal region granularity. The incorrect selection of region granularity is potentially prone to under- or over-segmentation of color images, which can lead to a non-uniform highlighting of salient objects. In this study, the method of color histogram clustering was utilized to automatically determine suitable homogenous regions in an image. Region saliency score was computed as a function of color contrast, contrast ratio, spatial feature, and center prior. Morphological operations were ultimately performed to eliminate the undesirable artifacts that may be present at the saliency detection stage. Thus, we have introduced a novel, simple, robust, and computationally efficient color histogram clustering method that agglutinates color contrast, contrast ratio, spatial feature, and center prior for detecting salient objects in color images. Experimental validation with different categories of images selected from eight benchmarked corpora has indicated that the proposed method outperforms 30 bottom-up non-deep learning and seven top-down deep learning salient object detection methods based on the standard performance metrics.
... We experiment on a 180 × 320 × 30 tensor which consists of the first 30 frames of the Sky dataset (available at http://www.loujing.com/rss-smalltarget, accessed on 28 July 2021) for small object detection [55]. ...
This paper conducts a rigorous analysis for the problem of robust tensor completion, which aims at recovering an unknown three-way tensor from incomplete observations corrupted by gross sparse outliers and small dense noises simultaneously due to various reasons such as sensor dead pixels, communication loss, electromagnetic interferences, cloud shadows, etc. To estimate the underlying tensor, a new penalized least squares estimator is first formulated by exploiting the low rankness of the signal tensor within the framework of tensor ∗L-Singular Value Decomposition (∗L-SVD) and leveraging the sparse structure of the outlier tensor. Then, an algorithm based on the Alternating Direction Method of Multipliers (ADMM) is designed to compute the estimator in an efficient way. Statistically, the non-asymptotic upper bound on the estimation error is established and further proved to be optimal (up to a log factor) in a minimax sense. Simulation studies on synthetic data demonstrate that the proposed error bound can predict the scaling behavior of the estimation error with problem parameters (i.e., tubal rank of the underlying tensor, sparsity of the outliers, and the number of uncorrupted observations). Both the effectiveness and efficiency of the proposed algorithm are evaluated through experiments for robust completion on seven different types of remote sensing data.
... The figure below shows that the 4 × 4 structure element is modelled with a chromosome length of 16 bits. For example, a chromosome equals [0,1,0,0;0,1,1,0;0,1,1,0;0,0,0,0], structural elements are shown as follow in Fig. 1. [19][20][21] α t i 2 represents the probability that the measured value of gene bit i in the t generation is 0 j i; β t i 2 , represents the probability that the measured value of gene bit i in the t generation is |1〉. The probability of setting each gene location to state "0" and state "1" is equal at the beginning, which can be taken as 2 ffiffi 2 p , This is shown below. ...
Robust point target detection of infrared clutter background has drawn great interest of scholars. Recently, morphological filter is playing a significant role in detecting infrared point target. Generally, the background clutter and targets are diverse in the case of each image. Traditional fixed structural elements and dimensions cannot be adjusted adaptively to acquire to successful point target detection in different complex backgrounds. Therefore, a new method is introduced based on quantum genetic algorithm to optimize and obtain structural element which is used as morphological filter for small target detection in original infrared images.Then,morphological contrast enhancement is further proposed to enhance energy of point targets after the filtered image is obtained.Thus, an enormous background clutter and noise are suppressed and the contrast between target and background are observably increased. Finally, by setting proper threshold, the point targets can be detected perfectly. Experimental evaluation results show that the proposed adaptive morphological contrast enhancement based on quantum genetic algorithm is effective and robust with respect to detection accuracy compared with the traditional morphological filter and other filtering algorithms.
... LSS-target detection in the visible light image and infrared images, respectively. Lou et al. [9] introduced the saliency and regional stability for feature extraction in visible-light images. A segmentation threshold was used to distinguish the target for accurate detection, but it was unsuitable for complex scenes. ...
Detection of the low-altitude, slow-speed, small (LSS) targets is one of the most popular research topics in remote sensing. Despite of a few existing approaches, there is still an accuracy gap for satisfying the practical needs. As the LSS targets are too small to extract useful features, deep learning based algorithms can hardly be used. To this end, we propose in this paper an effective strategy for determining the region of interest (ROI), using a multi-scale layered image fusion method to extract the most representative information for LSS target detection. In addition, an improved Self-Balanced Sensitivity Segment model (SuBSENSE) is proposed to detect the fused LSS-Target, which can further improve both the detection accuracy and the computational efficiency. We conduct extensive ablation studies to validate the efficacy of the proposed LSS target detection method on three public datasets and three self-collected datasets. The superior performance over the-state-of-the-arts (SOTA) has fully demonstrated the efficacy of the proposed approach.
... To demontrate the efficiency of our proposal, we compared the proposed method with various state of the art methods including image saliency methods (ITTI [7], GBVS [8], SUN [9], saliency by self-resemblance SSR [46], fast and efficient saliency (FES) [47], quaternion-based spectral saliency (QSS) [48], high-dimensional color transform (HDCT) [49], principle component analysis (PCA) [50], region stability saliency (RSS) [51]) and video saliency method ( consistent video saliency (CVS) [26], random walk with restart (RWRS) [27]), the implementation source code were collected from C.Wloka et al. [52] and the project page of the authors. We keep all parameters of the author's proposal as the default. ...
Intelligent surveillance systems enable secured visibility features in the smart city era. One of the major models for pre-processing in intelligent surveillance systems is known as saliency detection, which provides facilities for multiple tasks such as object detection, object segmentation, video coding, image re-targeting, image-quality assessment, and image compression. Traditional models focus on improving detection accuracy at the cost of high complexity. However, these models are computationally expensive for real-world systems. To cope with this issue, we propose a fast-motion saliency method for surveillance systems under various background conditions. Our method is derived from streaming dynamic mode decomposition (s-DMD), which is a powerful tool in data science. First, DMD computes a set of modes in a streaming manner to derive spatial–temporal features, and a raw saliency map is generated from the sparse reconstruction process. Second, the final saliency map is refined using a difference-of-Gaussians filter in the frequency domain. The effectiveness of the proposed method is validated on a standard benchmark dataset. The experimental results show that the proposed method achieves competitive accuracy with lower complexity than state-of-the-art methods, which satisfies requirements in real-time applications.
... Recently, there are also methods for directly detecting small targets in color images. For example, Jing Lou et al. [32] use five stability criteria such as area variation, center distance, and fill rate difference to generate a stability map. At the same time, they use the Gaussian blur method to generate a saliency map and then combine the stability map and the salient map for small target detection. ...
... Moreover, these targets do not have a fixed color distribution, and many of them tend to have a white distribution, which requires the designed saliency pathway cannot be based on a single color. Inspired by [32] in this paper, we turn the aerial image into Lab space. In order to suppress uniform regions and highlight small targets, we define standard deviations: ...
... (a) δ 1 , (b) δ 2 , (c) σ s (a) ground truth (b) RSS [19] (c) Our method Fig. 10 Test results for proposal generation methods. Compared to our method, the RSS [32] method generates smaller regions and is not complete enough the bounding box of the saliency region is used to directly crop the original image which always makes small targets incomplete in the cropped images. Therefore, we do not consider Method III in subsequent experiments. ...
Detecting small targets in large fields of view is a challenging task. Nowadays, many targets detection models based on the convolutional neural network (CNN) achieve excellent performance. However, these CNN-based detectors are inefficient when applied to tasks of real-time detection of small targets. This paper proposes a small-target detection model in large fields of view based on the tectofugal–thalamofugal–accessory optic system of birds. Within this model, first, we design an unsupervised saliency algorithm to generate saliency regions to suppress background information according to the visual information processing mechanism of the tectofugal pathway of birds. Second, we design a super-resolution (SR) analysis method to enlarge small targets and improve image resolution by the information processing mechanism of the accessory optic system of birds. Then, according to the information processing mechanism of the thalamofugal pathway, we propose a CNN-based method to detect small targets. We further test our model on two public datasets (the VEDAI dataset and DLR 3 K dataset), and the experimental results demonstrate that the proposed detection model outperforms the state-of-the-art methods on small-target detection.
... Cong et al. [14] designed a feature based on the depth confidence analysis and multiple cues fusion. Lou et al. [15] proposed an approach combining both regional stability and saliency for small road anomaly segmentation. ...
... Now we elaborate the RGB image processing pipeline inspired by [15]. The intuition behind the RGB processing pipeline is that the areas with different colors from surrounding areas are often marked as road anomalies. ...
The segmentation of drivable areas and road anomalies are critical capabilities to achieve autonomous navigation for robotic wheelchairs. The recent progress of semantic segmentation using deep learning techniques has presented effective results. However, the acquisition of large-scale datasets with hand-labeled ground truth is time-consuming and labor-intensive, making the deep learning-based methods often hard to implement in practice. We contribute to the solution of this problem for the task of drivable area and road anomaly segmentation by proposing a self-supervised learning approach. We develop a pipeline that can automatically generate segmentation labels for drivable areas and road anomalies. Then, we train RGB-D data-based semantic segmentation neural networks and get predicted labels. Experimental results show that our proposed automatic labeling pipeline achieves an impressive speed-up compared to manual labeling. In addition, our proposed self-supervised approach exhibits more robust and accurate results than the state-of-the-art traditional algorithms as well as the state-of-the-art self-supervised algorithms.
