Detection Of Salient Features Research Articles

Adopting multiple vision transformer layers for fine-grained image representation

Accurate discriminative regions proposal has an important effect for fine-grained image recognition. The vision transformer (ViT) brings about a striking effect in computer vision duo to its innate muti-head self-attention mechanism. However, the attention maps are gradually similar after certain layers and since ViT adds classification token for perform classification, it is unable to effectively select discriminative image patches for fine-grained image classification. To accurately detect discriminative regions, we propose a novel network AMTrans, which efficiently increases layers to learn diverse features and utilizes integrated raw attention maps to capture more salient feature. Specifically, we employ DeepViT as backbone to solve the attention collapse issue. Then, we fuse each head attention weight within each layer to produce attention weight map. After that, we alternatively use recurrent residual refinement blocks to promote salient feature detection and then utilize semantic grouping method to propose the discriminative feature region. A lot of experiments prove that AMTrans acquires the SOTA performance on three widely used fine-grained datasets under the same settings, involving Stanford-Cars, Stanford-Dogs and CUB-200-2011.

Multimodal brain image fusion based on error texture elimination and salient feature detection.

As an important clinically oriented information fusion technology, multimodal medical image fusion integrates useful information from different modal images into a comprehensive fused image. Nevertheless, existing methods routinely consider only energy information when fusing low-frequency or base layers, ignoring the fact that useful texture information may exist in pixels with lower energy values. Thus, erroneous textures may be introduced into the fusion results. To resolve this problem, we propose a novel multimodal brain image fusion algorithm based on error texture removal. A two-layer decomposition scheme is first implemented to generate the high- and low-frequency subbands. We propose a salient feature detection operator based on gradient difference and entropy. The proposed operator integrates the gradient difference and amount of information in the high-frequency subbands to effectively identify clearly detailed information. Subsequently, we detect the energy information of the low-frequency subband by utilizing the local phase feature of each pixel as the intensity measurement and using a random walk algorithm to detect the energy information. Finally, we propose a rolling guidance filtering iterative least-squares model to reconstruct the texture information in the low-frequency components. Through extensive experiments, we successfully demonstrate that the proposed algorithm outperforms some state-of-the-art methods. Our source code is publicly available at https://github.com/ixilai/ETEM.

Deep Learning Detection and Recognition of Spot Elevations on Historical Topographic Maps

Some information contained in historical topographic maps has yet to be captured digitally, which limits the ability to automatically query such data. For example, U.S. Geological Survey’s historical topographic map collection (HTMC) displays millions of spot elevations at locations that were carefully chosen to best represent the terrain at the time. Although research has attempted to reproduce these data points, it has proven inadequate to automatically detect and recognize spot elevations in the HTMC. We propose a deep learning workflow pretrained using large benchmark text datasets. To these datasets we add manually crafted training image/label pairs, and test how many are required to improve prediction accuracy. We find that the initial model, pretrained solely with benchmark data, fails to predict any HTMC spot elevations correctly, whereas the addition of just 50 custom image/label pairs increases the predictive ability by ∼50%, and the inclusion of 350 data pairs increased performance by ∼80%. Data augmentation in the form of rotation, scaling, and translation (offset) expanded the size and diversity of the training dataset and vastly improved recognition accuracy up to ∼95%. Visualization methods, such as heat map generation and salient feature detection, can be used to better understand why some predictions fail.

Image salient object detection with refined deep features via convolution neural network

Recent advances in saliency detection have used deep learning to obtain high-level features to detect salient regions. These advances have demonstrated superior results over previous works that use handcrafted low-level features for saliency detection. We propose a convolutional neural network (CNN) model to learn high-level features for saliency detection. Compared to other methods, our method presents two merits. First, when performing features extraction, apart from the convolution and pooling step in our method, we add restricted Boltzmann machine into the CNN framework to obtain more accurate features in intermediate step. Second, in order to avoid manual annotation data, we add deep belief network classifier at the end of this model to classify salient and nonsalient regions. Quantitative and qualitative experiments on three benchmark datasets demonstrate that our method performs favorably against the state-of-the-art methods.

Pedestrian recognition in multi-camera networks using multilevel important salient feature and multicategory incremental learning

The ability to recognize pedestrians across multiple camera views is of great importance for many applications in the broad field of video surveillance. Due to the absence of the topology and calibration of distributed cameras, spatio-temporal reasoning becomes unavailable, and therefore only appearance information can be used in real-world scenarios, especially for disjoint camera views. This paper proposes a novel approach based on important salient feature and multi-category transfer incremental learning to recognize pedestrians for long-term tracking in multi-camera networks without space-time cues. An accurate and robust model can be built for pedestrian recognition using few samples. We first propose a novel multi-level important salient feature detection method (MImSF11MImSF is the abbreviation of Multi-level Important Salient Feature detection method) to formulate the appearance model. Due to environmental changes, the appearances of the pedestrians under the camera can change over time and across space, therefore the classification performance may be impaired. Hence, the appearance models should be continuously updated. We then adopt a novel object recognition multicategory incremental modeling algorithm (ORMIM22ORMIM is the abbreviation of Object Recognition Multicategory Incremental Modeling algorithm) to update the appearance model adaptively and recognize the pedestrians based on a classification approach. One of the major advantages of the proposed method is that it can identify new target objects that were never learned in the primary model while improving the matching accuracy of what has been learned. We conduct extensive experiments on CAVIAR, ISCAPS databases and our own databases where the camera views are disjoint and the appearance of objects changes significantly due to variations in the camera viewpoint, illumination, weather and poses. The experiments demonstrate that our proposed model is superior to that of existing classification-based recognition methods in terms of accuracy, robustness and computation efficiency. The developed methodology can be used in retrieval, matching and other real-time video surveillance applications.

Presynaptic GABA Receptors Mediate Temporal Contrast Enhancement in Drosophila Olfactory Sensory Neurons and Modulate Odor-Driven Behavioral Kinetics.

Contrast enhancement mediated by lateral inhibition within the nervous system enhances the detection of salient features of visual and auditory stimuli, such as spatial and temporal edges. However, it remains unclear how mechanisms for temporal contrast enhancement in the olfactory system can enhance the detection of odor plume edges during navigation. To address this question, we delivered to Drosophila melanogaster flies pulses of high odor intensity that induce sustained peripheral responses in olfactory sensory neurons (OSNs). We use optical electrophysiology to directly measure electrical responses in presynaptic terminals and demonstrate that sustained peripheral responses are temporally sharpened by the combined activity of two types of inhibitory GABA receptors to generate contrast-enhanced voltage responses in central OSN axon terminals. Furthermore, we show how these GABA receptors modulate the time course of innate behavioral responses after odor pulse termination, demonstrating an important role for temporal contrast enhancement in odor-guided navigation.

Fully Automatic Detection of Salient Features in 3-D Transesophageal Images

Most automated segmentation approaches to the mitral valve and left ventricle in 3-D echocardiography require a manual initialization. In this article, we propose a fully automatic scheme to initialize a multicavity segmentation approach in 3-D transesophageal echocardiography by detecting the left ventricle long axis, the mitral valve and the aortic valve location. Our approach uses a probabilistic and structural tissue classification to find structures such as the mitral and aortic valves; the Hough transform for circles to find the center of the left ventricle; and multidimensional dynamic programming to find the best position for the left ventricle long axis. For accuracy and agreement assessment, the proposed method was evaluated in 19 patients with respect to manual landmarks and as initialization of a multicavity segmentation approach for the left ventricle, the right ventricle, the left atrium, the right atrium and the aorta. The segmentation results revealed no statistically significant differences between manual and automated initialization in a paired t-test (p > 0.05). Additionally, small biases between manual and automated initialization were detected in the Bland–Altman analysis (bias, variance) for the left ventricle (−0.04, 0.10); right ventricle (−0.07, 0.18); left atrium (−0.01, 0.03); right atrium (−0.04, 0.13); and aorta (−0.05, 0.14). These results indicate that the proposed approach provides robust and accurate detection to initialize a multicavity segmentation approach without any user interaction.

Multi-Frame Image Fusion Method Combining Spatial-Temporal Saliency Detection and NSCT

We propose a new image fusion method to fuse the frames of infrared and visual image sequences more effectively. In our method, we introduce an improved salient feature detection algorithm to achieve the saliency map of the original frames. This improved method can detect not only spatially but also temporally salient features using dynamic information of inter-frames. Images are then segmented into target regions and background regions based on saliency distribution. We formulate fusion rules for different regions using a double threshold method and finally fuse the image frames in NSCT multi-scale domain. Comparison of different methods shows that our result is a more effective one to stress salient features of target regions and maintain details of background regions from the original image sequences.

Distinct Mechanisms for Size Tuning in Primate Visual Cortex

Most neurons in primary visual cortex (V1) are selective for stimulus size, a property with important implications for salient feature detection. Size selectivity involves dynamic interactions between neuronal circuits that establish the classical (center) and extraclassical (surround) of a neuron's receptive field. Although much is known about the tuning properties and stimulus selectivity of the center and surround subunits, relatively little is known about how these subunits interact to achieve size selectivity. To address this question, we examined the temporal dynamics of size selectivity in two classes of pyramidal neurons at similar hierarchical processing stages in V1 of alert monkeys. These two classes were comprised of neurons in cortical layer 6 with identified projections to the lateral geniculate nucleus. While both neuronal groups displayed comparable levels of size selectivity, the temporal dynamics of their tuning differed significantly. We compared the size tuning profiles of each cell type with a series of sum-of-Gaussian models and discovered that the receptive fields of neurons with fast-conducting axons contained an excitatory center and a suppressive surround with similar onset timing. In contrast, neurons with slow-conducting axons used two center components-an early wide-field component and a delayed narrow-field component that increased activity-in addition to the surround component. The early, wide-field component represents a novel mechanism for cortical neurons to integrate contextual information. These results demonstrate that size tuning in cortical neurons is established via multiple unique mechanisms, dictated by the rich circuit architecture in which neurons are embedded.

Data Visualization: Featuring Interactive Visual Analysis

AbstractData visualization is an application‐driven field, that is always trying to satisfy its customers and to adapt to the demands, cultures, and workflows of many application areas. Therefore, it is difficult to keep focus on techniques and approaches that are not too application specific. A lot of good work on data visualization consists of single‐problem solutions, that cannot be easily merged into general‐purpose systems.In this talk, I will briefly review some current trends and issues, and identify some approaches that are common to many applications. One such approach in data visualization that has attracted interest from the early days is the detection of salient features, or patterns of interest in a data set. The main idea is to extract information at a higher level of abstraction from a mass of data, that is richer in semantics but much smaller in size, and that can help to define scenes and objects for visualization. This idea was pioneered in areas such as flow visualization, but is now more widely applied. It is often considered to be necessity to keep up with the ever rapidly increasing size of data sets, and the demand for interactivity in data visualization and analysis.Another generic approach in data visualization is called interactive visual analysis (TVA), consisting of a strongly interactive multiple‐linked‐view interfaces with integrated, powerful data analysis techniques taken from statistical analysis, pattern recognition, machine learning, and other fields. This is built on the assumptions that a single 2D or 3D visualization is often not enough, and spatial views can be augmented with abstract, derived data spaces; that strong interaction helps to promote insight; and that a better balance is needed between human visual inspection and computer‐based analysis and reasoning. Interestingly, an IVA interface can serve not only as an environment for exploration of low‐level data, but also for defining the high‐level features to be extracted, that should summarize the essence of the data. The high‐level features are usually highly application specific, and can only be found using theories from the application domains. The big challenge is to create environments for general purpose visual data analysis, and yet allow users to introduce advanced theories and methods from many application domains.The trend towards more integration in data visualization will be illustrated with cross‐links between very different areas, such as medical and flow visualization, and the combined use of techniques from scientific visualization and information visualization, and the absorption of other data analysis techniques. Also, historic and contemporary examples of feature extraction and interactive visual analysis will be shown.

Autonomous Salient Feature Detection through Salient Cues in an HSV Color Space for Visual Indoor Simultaneous Localization and Mapping

For successful simultaneous localization and mapping (SLAM), perception of the environment is important. This paper proposes a scheme to autonomously detect visual features that can be used as natural landmarks for indoor SLAM. First, features are roughly selected from the camera image through entropy maps that measure the level of randomness of pixel information. Then, the saliency of each pixel is computed by measuring the level of similarity between the selected features and the given image. In the saliency map, it is possible to distinguish the salient features from the background. The robot estimates its pose by using the detected features and builds a grid map of the unknown environment by using a range sensor. The feature positions are stored in the grid map. Experimental results show that the feature detection method proposed in this paper can autonomously detect features in unknown environments reasonably well.

Boosting color saliency in image feature detection

The aim of salient feature detection is to find distinctive local events in images. Salient features are generally determined from the local differential structure of images. They focus on the shape-saliency of the local neighborhood. The majority of these detectors are luminance-based, which has the disadvantage that the distinctiveness of the local color information is completely ignored in determining salient image features. To fully exploit the possibilities of salient point detection in color images, color distinctiveness should be taken into account in addition to shape distinctiveness. In this paper, color distinctiveness is explicitly incorporated into the design of saliency detection. The algorithm, called color saliency boosting, is based on an analysis of the statistics of color image derivatives. Color saliency boosting is designed as a generic method easily adaptable to existing feature detectors. Results show that substantial improvements in information content are acquired by targeting color salient features.

Interactive processing of sensory input and motor output in the human hippocampus.

Recent studies of visuomotor integration suggest that the motor system may be intimately involved in the detection of salient features of the sensory scene. The final stages of sensory processing occur in hippocampal structures. We measured human neuromagnetic responses during motor reaction to an auditory cue embedded in high-speed multimodal stimulation. Our results demonstrate that large-scale cognitive networks may recruit additional resources from the hippocampus during sensorimotor integration. Hippocampal activity from 300 msec before to 200 msec after cued movements was enhanced significantly over that observed during self-paced movements. The dominant hippocampal activity appeared equally synchronized to both sensory input and motor output, consistent with timing by an intrinsic mechanism, possibly provided by ongoing theta oscillations

Matching the resolution level to salient image features

This paper illustrates how strategies for structural description of images—based on the detection of salient features in the image—can be more efficiently realised by integrating them with the multiresolution approach to reduce details, up to the level at which such features are still recognisable. Since different structures of interest are visible in a restricted range of resolution levels, the combination of structural description with multi-resolution allows the automatic recognition of features at the minimum sufficient level, thus saving a costly description which would store irrelevant details.