Learning Discriminative Dictionary for Group Sparse Representation

To accomplish reliable pedestrian detection using unmanned aerial vehicles (UAVs) under night-time conditions, an image enhancement method is developed in this article to improve the low-illumination image quality. First, the image brightness is mapped to a desirable level by a hyperbolic tangent curve. Second, the block-matching and 3-D filtering methods are developed for an unsharp filter in YCbCr color space for image denoising and sharpening. Finally, pedestrian detection is performed using a convolutional neural network model to complete the surveillance task. Experimental results show that the Minkowski distance measurement index of enhanced images is increased to 0.975, and the detection accuracies, in F-measure and confidence coefficient, reach 0.907 and 0.840, respectively, which are the highest as compared with other image enhancement methods. This developed method has potential values for night-time UAV visual monitoring in smart city applications.

Low-Illumination Image Enhancement for Night-Time UAV Pedestrian Detection

Road scene perception of driving environment is an important yet stimulating task for an intelligent vehicle system (IVS). Advancement of deep learning techniques has enabled the image sensors to understand the scene more accurately for the intended object, especially in segmentation domain. In context of computer vision approach, segmentation of road scene from colour images is challenging, due to the varying illumination circumstances, the non-uniform road shapes and the fuzzy boundaries between the road and other objects. However, existing studies have limited performance in context of standard measures, therefore fails to incorporate the mentioned issues. With this aim and to clearly distinguish road and non-road portion, we have utilized U-Net, Seg-Net and fully convolutional network (FCN) models. From the conducted experiments, U-Net has achieved 94% score for mIoU and dice coefficient which is better than the scores achieved by other models Seg-Net and FCN-32. In our experiment, images from Camvid dataset have been used to train and validate the performance.

Road Detection Using Semantic Segmentation-Based Convolutional Neural Network for Intelligent Vehicle System

A multiscale and multidirectional network named the Contourlet convolutional neural network (CCNN) is proposed for synthetic aperture radar (SAR) image despeckling. SAR image resolution is not higher than that of optical images. If the network depth is increased blindly, the SAR image detail information flow will become quite weak, resulting in severe vanishing/exploding gradients. In this paper, a multiscale and multidirectional convolutional neural network is constructed, in which a single-stream structure of convolutional layers is replaced with a multiple-stream structure to extract image features with multidirectional and multiscale properties, thus significantly improving the despeckling performance. With the help of the Contourlet, the CCNN is designed with multiple independent subnetworks to respectively capture abstract features of an image in a certain frequency and direction band. The CCNN can increase the number of convolutional layers by increasing the number of subnetworks, which makes the CCNN not only have enough convolutional layers to capture the SAR image features, but also overcome the problem of vanishing/exploding gradients caused by deepening the networks. Extensive quantitative and qualitative evaluations of synthetic and real SAR images show the superiority of our proposed method over the state-of-the-art speckle reduction method.

Contourlet-CNN for SAR Image Despeckling

Robust road region extraction plays a crucial role in many computer vision applications, such as automated driving and traffic video analytics. Various weather and illumination conditions like snow, fog, dawn, daytime, and nighttime often pose serious challenges to automated road region detection. This paper presents a new real-time road recognition method that is able to accurately extract the road region in traffic videos under adverse weather and illumination conditions. Specifically, the novel global foreground modeling (GFM) method is first applied to subtract the ever-changing background in the traffic video frames and robustly detect the moving vehicles which are assumed to drive in the road region. The initial road samples are then obtained from the subtracted background model in the location of the moving vehicles. The integrated features extracted from both the grayscale and the RGB and HSV color spaces are further applied to construct a probability map based on the standardized Euclidean distance between the feature vectors. Finally, the robust road mask is derived by integrating the initially estimated road region and the regions located by the flood-fill algorithm. Experimental results using a dataset of real traffic videos demonstrate the feasibility of the proposed method for automated road recognition in real-time.

Robust Road Region Extraction in Video Under Various Illumination and Weather Conditions

In this paper, we propose a novel model of sparse representation for image denoising that we call an adaptive contourlet hidden Markov model (HMM)-pulse-coupled neural network (PCNN). In this study, we first adopted a contourlet transform to decompose a noisy image to be some subband coefficients of various directions at various scales. The contourlet emulated extremely well the sparse representation performance of human visual perception, such as its multiscale characteristics, geometric features, and bandpass properties. Second, we used an HMM method to create a statistical model that expressed the coefficient relationships in intrabands, interbands, intrascales, and interscales. Then we used an expectation-maximization training algorithm to obtain the state probability. The result included the state, scale, and direction, the position of the coefficient, the noisy image, and the parameter set of the HMM model. Third, we put the state probability into the PCNN model, which could adaptively optimize the parameters of the HMM model and get better coefficients of clean images. Finally, we transformed the image denoising problem into a Bayesian posterior probability estimation problem. We also reconstructed a denoised image based on the clean coefficients obtained from our proposed method. The experimental results show that the contourlet HMM-PCNN model proposed in this paper is superior to the contourlet with hidden Markov tree model and the wavelet threshold method.

/pdf/an-adaptive-contourlet-hmm-pcnn-model-of-sparse-zsxr92xpoq.pdf

An Adaptive Contourlet HMM–PCNN Model of Sparse Representation for Image Denoising

Multiscale geometric analysis (MGA) is not only characterized by multi-resolution, time-frequency localization, multidirectionality and anisotropy, but also outdoes the limitations of wavelet transform in representing high-dimensional singular data such as edges and contours. Therefore, researchers have been exploring new MGA-based image compression standards rather than the JPEG2000 standard. However, due to the difference in terms of the data structure, redundancy and decorrelation between wavelet and MGA, as well as the complexity of the coding scheme, so far, no definitive researches have been reported on the MGA-based image coding schemes. In addressing this problem, this paper proposes an image data compression approach using the hidden Markov model (HMM)/pulse-coupled neural network (PCNN) model in the contourlet domain. First, a sparse decomposition of an image was performed using a contourlet transform to obtain the coefficients that show the multiscale and multidirectional characteristics. An HMM was then adopted to establish links between coefficients in neighboring subbands of different levels and directions. An Expectation-Maximization (EM) algorithm was also adopted in training the HMM in order to estimate the state probability matrix, which maintains the same structure of the contourlet decomposition coefficients. In addition, each state probability can be classified by the PCNN based on the state probability distribution. Experimental results show that the HMM/PCNN -contourlet model proposed in this paper leads to better compression performance and offer a more flexible encoding scheme.

/pdf/a-combined-hmm-pcnn-model-in-the-contourlet-domain-for-image-8s2av3vc08.pdf

A combined HMM-PCNN model in the contourlet domain for image data compression.

Fast and robust vision-based road detection in an unstructured environment is very challenging. In this paper, we focus on vanishing-point (VP) detection in unstructured roads and propose a response-modulated line-voting method based on a contourlet transform, followed by a voter selection process for VP detection. We first adopt the contourlet transform to estimate the dominant vector for each pixel, including orientation and its relevant response. The estimated dominant vector is then selected by a novel select function to retrieve approximately 40% of the pixels with a reliable dominant vector in the image to vote. Unlike previous methods, this method takes into account the magnitudes of response of the pixels to improve the efficiency of the voting process by suppressing possible interference by extreme and strong textures. The pixels are given a moderate response to vote. Finally, for situations where the road texture is likely to be selected as a criterion for voting by the line-voting scheme, we use this simple and fast scheme to vote for the VP. We conduct experiments on a public dataset of 1,003 different types of natural road images as well as on our own dataset of 400 such images. The results demonstrate that in our dataset, the proposed method is comparable to and outperforms the state-of-the-art methods.

/pdf/fast-and-robust-vanishing-point-detection-using-contourlet-41ujs5gtm5.pdf

Fast and Robust Vanishing Point Detection Using Contourlet Texture Detector for Unstructured Road

This paper proposes a novel visual saliency model based on the CovSal algorithm of the region covariance matrices and histogram contrast (HC) method. First, we give a new CovSal algorithm of the local saliency contrast by improving the center-surrounder segmentation method. Second, we add the HC algorithm of the global saliency contrast, and then we sparsify the global saliency map using the low-rank matrix. Finally, we integrate the local and global saliency maps through combining both the CovSal and HC contrast. Present paper proposes a new visual saliency model that combines the local and global saliency contrast algorithms, simultaneously retaining their advantages and eliminating their drawbacks. For example, the proposed model reduces the influence of background and texture details. The experimental results show that the performance of the visual saliency model proposed in this paper has been improved compared to the CovSal algorithm presented by Erdem and Erdem.

Research on Visual Saliency Model Based on CovSal Algorithm and Histogram Contrast

The invention discloses a road vanishing point detection method for an unstructured road single image. The method comprises the following steps: 1, adjusting the size of an input road image; 2, extracting a texture vector set of the road image through contourlet decomposition for the adjusted image, and determining a main texture vector; 3, screening the pixel points through the extracted texturevector set and the extracted main texture vector to obtain a voting point set; 4, modulating texture amplitudes of pixel points in the voting point set; and 5, voting according to a line voting methodto detect road vanishing points. According to the vanishing point detection method provided by the invention, the accuracy of vanishing point detection is effectively improved on the premise of ensuring relatively good real-time performance, and particularly in roads with fuzzy textures or complex and multi-directional textures, the accuracy is higher than that of previous methods.

Yuhao Wang

Papers

An Adaptive Contourlet HMM–PCNN Model of Sparse Representation for Image Denoising

A combined HMM-PCNN model in the contourlet domain for image data compression.

Fast and Robust Vanishing Point Detection Using Contourlet Texture Detector for Unstructured Road

Research on Visual Saliency Model Based on CovSal Algorithm and Histogram Contrast

Road vanishing point detection method for unstructured road single image