Performance Analysis of Otsu-Based Thresholding Algorithms: A Comparative Study

Image thresholding is a widely used technology for a lot of computer vision applications, and among various global thresholding algorithms, Otsu-based approaches are very popular due to their simplicity and effectiveness. While the usage of Otsu-based thresholding methods is well discussed, the performance analyses of these methods are rather limited. In this paper, we first review nine Otsu-based approaches and categorize them based on their objective functions, preprocessing, and postprocessing strategies. Second, we conduct several experiments to analyze the model characteristics using different scene parameters both on synthetic images and real-world cell images. We put more attention to examine the variance of foreground object and the effect of the distance between mean values of foreground and background. Third, we explore the robustness of algorithms by introducing two typical kinds of noises under different intensities and compare the running time of each method. Experimental results show that NVE, WOV, and Xing’s methods are more robust to the distance of mean values of foreground and background. The large foreground variance will cause a larger threshold value. Experiments on cell images show that foreground miss detection becomes serious when the intensities of foreground pixels change drastically. We conclude that almost all algorithms are significantly affected by Salt&Pepper and Gaussian noises. Interestingly, we find that ME increases almost linearly with the intensity of Salt&Pepper noise. In terms of algorithms’ time cost, methods with no preprocessing and postprocessing steps have more advantages. All these findings can serve as a guideline for image thresholding when using Otsu-based thresholding approaches.

Semantic Segmentation of Remote Sensing Image Based on Convolutional Neural Network and Mask Generation

High-resolution remote sensing images usually contain complex semantic information and confusing targets, so their semantic segmentation is an important and challenging task. To resolve the problem of inadequate utilization of multilayer features by existing methods, a semantic segmentation method for remote sensing images based on convolutional neural network and mask generation is proposed. In this method, the boundary box is used as the initial foreground segmentation profile, and the edge information of the foreground object is obtained by using the multilayer feature of the convolutional neural network. In order to obtain the rough object segmentation mask, the general shape and position of the foreground object are estimated by using the high-level features in the process of layer-by-layer iteration. Then, based on the obtained rough mask, the mask is updated layer by layer using the neural network characteristics to obtain a more accurate mask. In order to solve the difficulty of deep neural network training and the problem of degeneration after convergence, a framework based on residual learning was adopted, which can simplify the training of those very deep networks and improve the accuracy of the network. For comparison with other advanced algorithms, the proposed algorithm was tested on the Potsdam and Vaihingen datasets. Experimental results show that, compared with other algorithms, the algorithm in this article can effectively improve the overall precision of semantic segmentation of high-resolution remote sensing images and shorten the overall training time and segmentation time.

A Framework for Multi-Threshold Image Segmentation of Low Contrast Medical Images

Accurate medical images segmentation plays a vital role in contouring during diagnosis and treatment planning. To improve the segmentation accuracy in low contrast images, we propose a method by combining Hill entropy and fuzzy c-partition. Here, using membership function, an image is first transformed into fuzzy domain. Subsequently, the fuzzy Hill entropies are defined for foreground (object) and background. Next, the total fuzzy Hill entropy is maximized to compute the accurate threshold; this process is employed to calculate a proper parameter combination of membership function. This Hill entropy is then optimized to acquire an image threshold by Differential Evolution “DE” optimization algorithm. The key benefit of the presented approach is that it considers the information of background and object as well as interactions between them in threshold selection mechanism. The results and performance evaluations show the better accuracy of our technique over other existing approaches.

Analysis of spiking synchrony in visual cortex reveals distinct types of top-down modulation signals for spatial and object-based attention

The activity of a border ownership selective (BOS) neuron indicates where a foreground object is located relative to its (classical) receptive field (RF). A population of BOS neurons thus provides an important component of perceptual grouping, the organization of the visual scene into objects. In previous theoretical work, it has been suggested that this grouping mechanism is implemented by a population of dedicated grouping (“G”) cells that integrate the activity of the distributed feature cells representing an object and, by feedback, modulate the same cells, thus making them border ownership selective. The feedback modulation by G cells is thought to also provide the mechanism for object-based attention. A recent modeling study showed that modulatory common feedback, implemented by synapses with N-methyl-D-aspartate (NMDA)-type glutamate receptors, accounts for the experimentally observed synchrony in spike trains of BOS neurons and the shape of cross-correlations between them, including its dependence on the attentional state. However, that study was limited to pairs of BOS neurons with consistent border ownership preferences, defined as two neurons tuned to respond to the same visual object, in which attention decreases synchrony. But attention has also been shown to increase synchrony in neurons with inconsistent border ownership selectivity. Here we extend the computational model from the previous study to fully understand these effects of attention. We postulate the existence of a second type of G-cell that represents spatial attention by modulating the activity of all BOS cells in a spatially defined area. Simulations of this model show that a combination of spatial and object-based mechanisms fully accounts for the observed pattern of synchrony between BOS neurons. Our results suggest that modulatory feedback from G-cells may underlie both spatial and object-based attention.

The forest, the trees, or both? Time-course and interactions between gist and object processing during natural scenes perception

The global-to-local theories of perception assume that the gist of a scene is computed early and automatically, whereas objects are recognized at later stages and in relation to the gist. To test these theoretical predictions we investigated the time course of gist- and object-recognition, and their interaction in two experiments (total N = 60). Scene images consisting of a background and a foreground object were displayed briefly (between 8 ms and 100 ms) and backward masked. As expected, we found better categorization of backgrounds than objects, and impaired categorization of objects semantically incongruent with backgrounds. But we also observed impaired recognition of backgrounds when an incongruent object was present in the scene, an effect not predicted theoretically. Therefore, we confirm that gist is recognized first and affects subsequent object recognition, but the fact that objects also influence gist processing suggests a bidirectional interaction between both processes.

Privacy-Preserving Surveillance as an Edge Service Based on Lightweight Video Protection Schemes Using Face De-Identification and Window Masking

With a myriad of edge cameras deployed in urban and suburban areas, many people are seriously concerned about the constant invasion of their privacy. There is a mounting pressure from the public to make the cameras privacy-conscious. This paper proposes a Privacy-preserving Surveillance as an Edge service (PriSE) method with a hybrid architecture comprising a lightweight foreground object scanner and a video protection scheme that operates on edge cameras and fog/cloud-based models to detect privacy attributes like windows, faces, and perpetrators. The Reversible Chaotic Masking (ReCAM) scheme is designed to ensure an end-to-end privacy while the simplified foreground-object detector helps reduce resource consumption by discarding frames containing only background-objects. A robust window-object detector was developed to prevent peeping via windows; whereas human faces are detected by using a multi-tasked cascaded convolutional neural network (MTCNN) to ensure de-identification. The extensive experimental studies and comparative analysis show that the PriSE scheme (i) can efficiently detect foreground objects, and scramble those frames that contain foreground objects at the edge cameras, and (ii) detect and denature window and face objects, and identify perpetrators at a fog/cloud server to prevent unauthorized viewing via windows, to ensure anonymity of individuals, and to deter criminal activities, respectively.