Visibility Estimation Based on Weakly Supervised Learning under Discrete Label Distribution.

This paper proposes an end-to-end neural network model that fully utilizes the characteristic of uneven fog distribution to estimate visibility in fog images. Firstly, we transform the original single labels into discrete label distributions and introduce discrete label distribution learning on top of the existing classification networks to learn the difference in visibility information among different regions of an image. Then, we employ the bilinear attention pooling module to find the farthest visible region of fog in the image, which is incorporated into an attention-based branch. Finally, we conduct a cascaded fusion of the features extracted from the attention-based branch and the base branch. Extensive experimental results on a real highway dataset and a publicly available synthetic road dataset confirm the effectiveness of the proposed method, which has low annotation requirements, good robustness, and broad application space.

VISOR-NET: Visibility Estimation Based on Deep Ordinal Relative Learning under Discrete-Level Labels.

This paper proposes a novel end-to-end pipeline that uses the ordinal information and relative relation of images for visibility estimation (VISOR-NET). By encoding ordinal information into a set of relatively ordered image pairs, VISOR-NET can learn a global ranking function effectively. Due to the lack of real scenes or continuous labels in public foggy datasets, we collect a large-scale dataset that we term Foggy Highway Visibility Images (FHVI), which are taken from real surveillance scenes, and synthesize an INDoor Foggy images dataset (INDF) with continuous annotation. This work measures the estimation effectiveness on two public datasets and our FHVI dataset as a classification task and then on the INDF dataset as a regression task. Comprehensive experiments with existing deep-learning methods demonstrate the performance of the proposed method in terms of estimation accuracy, the convergence rate, model stability, and data requirements. Moreover, this method can extend inter-level visibility estimation to intra-level visibility estimation and can realize approximate regression estimation under discrete-level labels.

A convolutional neural network Cascade for plantar pressure images registration.

BACKGROUND: Plantar pressure image (PPI) recorded in high spatial and temporal resolution is very useful in clinical gait analysis. For functional analysis of PPI, image registration is often performed to maximally correlate source image with a template image. Previous methods estimate the registration parameters by iteratively optimizing different objective functions. These methods are often computational expensive to achieve satisfactory registration accuracy. RESEARCH QUESTION: Can we develop a single PPI registration technique that performs more rapidly than previous methods, and that also maintains adequate PPI correspondence as defined by various (dis)similarity metrics? METHODS: A cascaded convolutional neural network (CNN) was proposed for the registration of PPIs. Our model was trained to learn a regression from the difference between the template and misaligned images to the registration parameters. The registration performance was evaluated by three different metrics, i.e. the mean squared error (MSE), the exclusive or (XOR), and the mutual information (MI). For comparison, four previous methods were also implemented. These included the principal axes (PA) method, the center of pressure trajectory (COP) method, the MSE method, and the XOR method. RESULTS: Experimental results on a dataset with 71 PPI template-source pairs showed that the proposed CNN-based method could obtain comparable registration accuracy to the MSE and XOR method. With regards to the registration speed, registration durations (mean ± sd in seconds) per image pair were: MSE (30.584 ± 2.171), XOR (24.245 ± 1.596), PA (0.016 ± 0.003), COP (25.614 ± 0.341), and the proposed model (0.054 ± 0.007). SIGNIFICANCE: Our findings indicate that the proposed registration approach can achieve high accuracy but less computational time. Thus, it is more practical to utilize our pre-trained CNN-based model to develop near-real time applications for plantar pressure images registration.

Approximate sparse spectral clustering based on local information maintenance for hyperspectral image classification.

Sparse spectral clustering (SSC) has become one of the most popular clustering approaches in recent years. However, its high computational complexity prevents its application to large-scale datasets such as hyperspectral images (HSIs). In this paper, we propose two efficient approximate sparse spectral clustering methods for HSIs clustering in which clustering performance is improved by utilizing local information among the data. Firstly, we construct a smaller representative dataset on which sparse spectral clustering is performed. Then the labels of ground object are extending to whole dataset based on the local information according to two extending strategies. The first one is that the local interpolation is utilized to improve the extension of the clustering result. The other one is that the label extension is turned to a problem of subspace embedding, and is fulfilled by locally linear embedding (LLE). Several experiments on HSIs demonstrated that the proposed algorithms are effective for HSIs clustering.