Salient object detection in video using deep non-local neural networks

The goal of this special issue was to bring together experts working on applying models, principles, and knowledge of human audio-visual perception and cognition to optimize video processing algorithms and applications. This is an area where our deeper understanding of visual perception to video processing will lead to new breakthroughs in visual information processing. Attentional focus is an important aspect of perception that helps understand user interest and intent. Determining points of saliency is computationally complex. Compressed-domain methods thus become valuable tools for developing computationally efficient and practical solutions. In BCompressed-Domain Correlates of Human Fixations in Dynamic Scenes^ (10. 1007/s11042-015-2802-3) authors present a method for detecting points of fixation in H.264/ AVC video using motion vectors, block coding modes and coded residuals parsed from a H. 164/AVC bitstream. Egocentric videos are captured usingwearable cameras and used to detect a wearer’s point of view. Amount of video captured using wearable cameras is increasing and saliency detection in such videos enables applications such as efficient video summarization. In BGeometrical Cues in Visual Saliency Models for Active Object Recognition in Egocentric Videos^ (10.1007/ s11042-015-2803-2) authors use geometrical cues to improve saliency detection in videos. Action recognition from videos is a challenging task with many applications including surveillance and social behavior understanding. Inspired by models of neural response to visual input, the paper entitled BDeep Learning Human Actions from Video via Sparse Filtering and Locally Competitive Algorithms^ (10.1007/s11042-015-2808-x) presents a method that combines sparse filtering with locally competitive algorithms for applications in action recognition. Multimed Tools Appl (2015) 74:10053–10056 DOI 10.1007/s11042-015-2957-y

Salient object detection in video has attracted enormous research efforts for its wide applicability. But there are still some issues in restraining the disturbance of background, which make it difficult to detect salient object in complex scenarios. Inspired by the hypothesis of center prior in image domain, we novelly introduced the concept of dynamic attention center in video. The distance between specific regions and this center is used as a weighting term to restrain the influence of background disturbance and obtain more accurate spatial and temporal saliency maps. Besides, we develop a dynamic fusion method to combine the temporal and spatial saliency map, leading to higher spatiotemporal consistency. The experiments on Freiburg-Berkeley Motion Segmentation Dataset show that our method outperforms several state-of-art methods on subjective visual perception and objective measurements.

Neurobiology researches suggest that the motion information attracts more attention of human visual system than other low-level features such as brightness, color and texture. Consequently, video saliency detection methods not only consider the spatial saliency caused by the underlying features of images, but also the motion information in temporal domain. In this study, we proposes a model of video salient object detection based on a two-pathway framework that the spatio-temporal contrast guides the search for salient targets. Firstly, along the non-selective pathway, which is computed with the intra-frame and inter-frame maps of the color contrast and motion contrast, combining with the previous saliency map, to represent the prior information of the possible target locations. In contrast, the low-level features such as brightness, color and motion features are extracted in the selective pathway to search target accurately. Finally, the Bayesian inference is used to further obtain the optimal results. Experimental results show that our algorithm improves the performance of salient object detection on video compared to the representative method of Contour Guided Visual Search.

Nowadays, salient object detection (SOD) has become a prominent research area in computer vision, which has various applications in real life. With the advancement of convolutional neural networks (CNN), numerous SOD methods have been proposed to imitate the human visual system to identify salient objects in images. Their marvelous performance comes with large models, high computational complexity, and memory consumption, raising many challenges to deploy these models on practical applications and low-resource devices. Under the high demand for developing efficient models for SOD tasks, in this paper we provide a comprehensive performance analysis focusing on existing CNN-based lightweight SOD models in terms of accuracy and efficiency. By discussing recent methods regarding the proposed ideas and main techniques, we analyze the current issues of lightweight SOD models in detail. Various metrics are considered for performance assessment followed by future research directions and conclusion.

Nowadays, ship detection in sea-sky background is not only useful in maritime visual surveillance, but also helpful in maritime search and rescue. Since ships are salient objects in infrared images with sea-sky background, we present a novel and effective algorithm based on saliency for ship detection in this situation. Our algorithm adopts global saliency, local saliency and background prior to generate saliency maps. Ships are finally segmented in saliency maps. Our algorithm is compared with four classic salient object detection algorithms. And experimental results show our algorithm outperforms the other four algorithms in qualitative and quantitative terms. Keyword list: ship detection ; salient object detection ; infrared image ; sea - sky background . 1. INTRODUCTION Since ships are important vehicles for sea transportation, ship detection in sea-sky background is extensively used in maritime visual surveillance and reconnaissance, avoidance of ship collision accidents, maritime search and rescue, etc. Infrared imaging systems can almost work in any environment and all-weather conditions with its visual capacity in darkness and frog, which makes it a very suitable tool in this application. In recent years, ship detection in infrared images with sea-sky background attracts increasing attention. Many methods are proposed in this field, including image segmentation methods [1-2] and morphological methods [3]. However, due to complex sea-sky environment and large variation of ship sizes, these methods cannot solve the issue completely. And since ships are often salient objects in infrared images with sea-sky background, some researchers have recently tried salient object detection ideas used in color images [4-6]. And the paper is focused on using salient object detection methods to solve the issue. Salient objects are objects which stand out in the image and attract most people ¶VLPPHGLDWHDWWHQWLRQ7KHVWDQGDUGway to detect salient ob MHFWVLVILUVWO\XVLQJVDOLHQF\GHWHFWLRQPHWKRGVWRJHQHUDWHVDOLHQF\PDSVLQZKLFKDSL[HO¶VJUH\value is proportional to its degree of saliency and then segmenting salient objects in saliency maps. Many salient detection methods can be divided into three categories: local contrast methods, global rarity methods and other methods. In local contrast methods, regions which have high contrast with local neighbouring regions are regarded as salient regions. Itti et al. [7] decompose an image into several feature maps . The saliency in each feature map is measured by local center-surround difference in several scales and saliency of different feature maps are integrated to a final saliency

Perceptual localization and focus refinement network for RGB-D salient object detection

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.

Detecting multiple salient objects in complex scenes is a challenging task. In this paper, we present a novel method to detect salient objects in images. The proposed method is based on the general ‘center-surround’ visual attention mechanism and the spatial frequency response of the human visual system (HVS). The saliency computation is performed in a statistical way. This method is modeled following three biologically inspired principles and compute saliency by two ‘scatter matrices’ which are used to measure the variability within and between two classes, i.e., the center and surrounding regions, respectively. In order to detect multiple salient objects of different sizes in a scene, the saliency of a pixel is estimated via its saliency support region which is defined as the most salient region centered at the pixel. Compliance with human perceptual characteristics enables the proposed method to detect salient objects in complex scenes and predict human fixations. Experimental results on three eye tracking datasets verify the effectiveness of the method and show that the proposed method outperforms the state-of-the-art methods on the visual saliency detection task.

With the rapid development of depth sensor, more and more RGB-D videos could be obtained. Identifying the foreground in RGB-D videos is a fundamental and important task. However, the existing salient object detection (SOD) works only focus on either static RGB-D images or RGB videos, ignoring the collaborating of RGB-D and video information. In this paper, we first collect a new annotated RGB-D video SOD (ViDSOD-100) dataset, which contains 100 videos within a total of 9,362 frames, acquired from diverse natural scenes. All the frames in each video are manually annotated to a highquality saliency annotation. Moreover, we propose a new baseline model, named attentive triple-fusion network (ATF-Net), for RGB-D video salient object detection. Our method aggregates the appearance information from an input RGB image, spatio-temporal information from an estimated motion map, and the geometry information from the depth map by devising three modality-specific branches and a multi-modality integration branch. The modality-specific branches extract the representation of different inputs, while the multi-modality integration branch combines the multi-level modalityspecific features by introducing the encoder feature aggregation (MEA) modules and decoder feature aggregation (MDA) modules. The experimental findings conducted on both our newly introduced ViDSOD-100 dataset and the well-established DAVSOD dataset highlight the superior performance of the proposed ATF-Net.This performance enhancement is demonstrated both quantitatively and qualitatively, surpassing the capabilities of current state-of-the-art techniques across various domains, including RGB-D saliency detection, video saliency detection, and video object segmentation. Our data and our code are available at ViDSOD-100.

Deep (machine) learning in recent years has significantly increased the predictive modeling strengths for applications in the areas of computer vision, speech recognition, and natural language processing. Being heavily data-driven, conventional general-purpose deep learning methods mainly rely on training data to learn task-specific data representations. In this thesis, we focus on leveraging novel learning algorithms in deep learning to tackle visual recognition challenges at the three key levels -- namely pixel, object, and image levels. Throughout the thesis, we demonstrate how each and every of our proposed methods can address a specific visual recognition problem. Firstly, we apply deep learning to improve the saliency detection performance of deep networks. A coarse saliency map of the input image's salient object is initially predicted by a convolutional-deconvolutional network (or equivalently a fully convolutional network). Following that, the coarse saliency map is sequentially refined by a separate network called Recurrent Attentional Convolutional-Deconvolutional Network (RACDNN) that shares its network weights across its sequential steps. At every sequential (temporal) step, the same RACDNN is applied to a newly selected/attended image sub-region and it performs saliency map refinement for that particular sub-region. Despite having many learnable network parameters and no access to additional training data, RADCNN can be trained well and it outperforms the "single-step" (coarse saliency detection) baseline and state-of-the-art saliency detection methods by large margins, on several saliency detection benchmark datasets. Secondly, we introduce auto-encoder weight transfer networks (AE-WTN+), a deep model for scaling up object detection. Conventional WTN transfers object class knowledge from the classification weights of a pretrained large-scale image classification network, to the classification weights of an object detection network. However, they suffer from under-fitting and are not able to preserve the rich object class information of novel classes. In this work, we propose a new WTN method -- AE-WTN+ which makes use of normalization techniques to ease network optimization. Inputs and intermediate features are normalized to improve trainability for reduced under-fitting. More importantly, a novel auto-encoding (reconstruction) loss is proposed to encourage information preservation of all classes in the weight predictions of AE-WTN+. This enables the predicted weights of AE-WTN+ to maintain the class discriminability and neighbourhood relationships contained in the pre-trained large-scale classification weights. Experiments on large-scale detection datasets validate the effectiveness of AE-WTN+ on both seen and novel classes. Thirdly, we improve the efficiency of cross-layer network connections for very deep convolutional network in the context of image classification. This gives rise to a new convolutional network architecture called Deluge Networks (DelugeNets) which pass the output (feature activations) of any preceding composite network layer as direct inputs to all succeeding composite layers. For the incoming feature activations of preceding composite layers, the succeeding composite layers first apply rather lightweight cross-layer depthwise convolutional weights on them for feature aggregation. The heavy lifting is carried out by the last convolutional weight layer in each preceding composite layer that performs cross-channel convolutions. The proposed technique enables unimpeded information flow among many layers in very deep DelugeNets with great efficiency. Experiments on small-scale and large-scale datasets show that DelugeNets perform competitively to state-of-the-art deep convolutional networks at lower computational and parameter costs. Fourthly, we propose Stochastic Downsampling (SDPoint), a method to overcome a major limitation suffered by conventional deep convolutional networks -- their computational costs during inference are fixed and cannot be changed. SDPoint downsamples the feature maps at a randomly selected point (layer index) in the network hierarchy during training. The many downsampling configurations known as SDPoint instances each entail a unique downsampling point and downsampling ratio. The same convolutional network weights are shared across the different SDPoint instances that downsample feature maps differently. During inference, the computational cost of a network trained with SDPoint can be instantaneously adjusted to fit a given computational budget, by manually selecting an appropriate SDPoint instance. Additionally, sharing weights between different feature map scales provides significant regularization benefits, making the networks trained with SDPoint less sensitive to scales and more competent. In image classification experiments, SDPoint demonstrates significant cost-accuracy and computational advantages, over independently-trained (non-weight-shared) baselines with various inference costs.

Image saliency detection has recently witnessed significant progress due to deep convolutional neural networks. However, extending state-of-the-art saliency detectors from image to video is challenging. The performance of salient object detection suffers from object or camera motion and the dramatic change of the appearance contrast in videos. In this paper, we present flow guided recurrent neural encoder (FGRNE), an accurate and end-to-end learning framework for video salient object detection. It works by enhancing the temporal coherence of the per-frame feature by exploiting both motion information in terms of optical flow and sequential feature evolution encoding in terms of LSTM networks. It can be considered as a universal framework to extend any FCN based static saliency detector to video salient object detection. Intensive experimental results verify the effectiveness of each part of FGRNE and confirm that our proposed method significantly outperforms state-of-the-art methods on the public benchmarks of DAVIS and FBMS.

Salient object detection has drawn much attention recently. The extracted salient objects can be used for tasks including recognition, segmentation and retrieval. Most existing models for saliency detection can only be applied to the high contrast and normal contrast scenes. They are not robust enough to handle objects in low contrast images. Due to low lightness, low Signal to Noise Ratio (SNR) and low appearance information, saliency detection in low contrast scenes remains a challenge. To solve this problem, this paper proposes a saliency detection model based on the feature extraction, in which optimal features have been learned in different contrast environment to improve the effectiveness of salient object detection. Compared with other existing saliency models, the proposed method has strong adaptability for different scenes, and gives superior performance in detecting the salient object especially in low contrast images.

Saliency detection for images and videos becomes increasingly popular due to its wide applicability. Enormous research efforts have been focused on saliency detection, but it still has some issues in maintaining spatiotemporal consistency of videos and uniformly highlighting entire objects. To address these issues, this paper proposes a superpixel-level spatiotemporal saliency model for saliency detection in videos. To detect salient object, we extract multiple spatiotemporal features combined with intra-consistency motion information preliminarily. Meanwhile, considering inter-consistency of foreground in videos, a set of foreground locations are obtained from previous frames. Then, we introduce foreground-background and local foreground contrast saliency cues of those features using the location prior information of foreground. These two improved contrast saliency cues uniformly highlight the entire object and suppress the background effectively. Finally, we use an interactively dynamic fusion method to integrate the output spatial and temporal saliency maps. The proposed approach is validated on challenging sets of video sequences. Subjective observations and objective evaluations demonstrate that the proposed model achieves a better performance on saliency detection compared with the state-of-the-art spatiotemporal saliency methods.

During past few years’ significant usage of video monitoring has been observed in several places. Video monitoring has always helped in improvement of security specific aspects. These systems provide major helping hand in monitoring through multiple object detection methods. However, there are certain challenges related to weather conditions which needs to be addressed. In view of this in this paper, we study multiple object detection in surveillance videos through hybrid deep gaussian mixture ensembles. This method integrates multiple deep gaussian mixture components to form ensembles which helps to develop multiple object detection solutions in surveillance videos. The computational system consists of ensemble of three major deep gaussian mixture components. The ensemble pipeline is formed with augmenting, bagging and stacking methods. The ensemble consists of hybrid object detection paradigm with steps background analysis, video pre-processing, data integration, object detection, foreground detection and post-processing. The computational pipeline is successfully experimented with ViSOR and CDnet 2014 benchmarked datasets. All results are validated with accuracy, precision, sensitivity or recall, specificity, F1-Score and RMSE metrics. Several comparative studies are performed with state-of-the-art methods as well as baselines. The experimental results demonstrate superiority of this method in comparison with other methods. Ablation studies have also been performed here with superior results. This system presents robustness in multiple object detection for real life challenges such as sudden illumination variation, shadow presence, long term occlusion and formation of dynamic backgrounds. It provides cost-effective, more profitable, efficient and sustainable real time multiple object detection solutions

Moving objects detection in video is essential for many computer vision applications and it is considered as a challenging research issue due to dynamic changes in object size, shape, complex background and illumination changes. In this research article a novel method to detect moving objects in video is proposed grounded on Locality Preserving Projections (LPP) and also thorough analysis of variations in neighbor mode and weight mode for constructing adjacency graph is given. LPP is an unsupervised subspace learning approach used for dimensionality reduction, which preserves the neighborhood structure of dataset while detection, which is vital in further steps accurate tracking and recognition of objects. The proposed method is tested on standard datasets with complex environments and experimental results obtained with variations in neighborhood modes and weight modes are encouraging.