2017 2nd International Conference on Image, Vision and Computing (ICIVC)

We present an efficient technique for high-definition image and video salient object detection using a hierarchical Boolean map approach. We begin by extracting multiple boolean map layers. Within each layer, we then apply flood fill algorithm to each seed pixel in parallel to generate attention maps. The saliency map is calculated by summing up all the attention maps. We further improve video con... View full abstract»

Saliency detection becomes a crucial requirement for numerous computer vison application. Conventional manifold ranking models have been widely used for saliency detection because it can measure similarity efficiently between the regions, but most of them make use of color and texture information and location information of objects didn't be well exploited, so it cannot work properly when objects ... View full abstract»

Aiming at the problem of low ship detection performance of SAR image in port area, this paper proposes SAR image ship detection based on sea clutter accurate modeling in port area. In order to establish the sea clutter model accurately, this paper tries to eliminate all the pixels of land and the target of the suspected ships before modeling. Firstly, the SAR image is preprocessed by fine Lee rate... View full abstract»

Automatic airport detection has received great attention due to the importance of airports in both military and civilian uses. This paper focuses on automatic airport detection in large-size remote sensing images under a two-step object detection framework. In the first step, both geometrical saliency and local entropy saliency are improved to find more accurate ROIs for detecting airports in larg... View full abstract»

This paper presents a novel scene text detection algorithm based on Stroke Width Transform (SWT), Maximally Extremal Regions (MSER) and candidate classification. Firstly, utilize the SWT and MSER to extract the candidate characters at the same time. Secondly, preliminary filtering the candidate connected components based on heuristic rules. Thirdly, using mutual verification and integration to cla... View full abstract»

In this paper, a new method of saliency-based traffic sign detection is presented. On the basis of the visual attention mechanism model, edge and color information are extracted as early visual features, and each feature is computed and normalized to obtain feature maps, conspicuity maps and the saliency map. Then the candidate regions containing traffic signs are determined with self-organizing m... View full abstract»

Since most energetic celestial sources in the Universe often exhibit a point-like appearance in the observed images, design of methods to detect and extract them is a hot research topic. An approach to detect and extract point sources in X-ray astronomical images that uses Gaussian smooth filters is proposed. As there might be false detections, a spurious source removal approach taking advantage o... View full abstract»

During the pharmaceutical process, it is inevitable that various defects emerge in the medicine vials which may greatly affect the product quality and reduce the productive efficiency. To address these problems, a method based on feature extraction and machine learning is developed for vial defect inspection. On image preprocessing, we used threshold algorithm to acquire the region of interest (RO... View full abstract»

With the popularity of surveillance video, face detection in surveillance video has become a popular and important topic. Face detection in surveillance video plays an important role in many popular applications such as: personal identification, crowd analysis, database establishment, and abnormal event detection. This paper proposes an unconstrained face detection method for surveillance video, w... View full abstract»

In this paper, a new method based on background prediction and high-order statistics for infrared dim small target detection is proposed. Firstly, the wavelet filter is introduced to remove the target on image as noise, which could efficiently estimate the distribution of the background image. Secondly, the candidate target components are extracted from the foreground image. Finally, high-order st... View full abstract»

Automatic target recognition is a challenging task as the response from an underwater target may vary greatly depending on its configuration, sonar parameters and the environment. In forward looking sonar image the target is considered as composed of a few rows and columns of highlight pixels and a few rows and some columns of shadow pixels. We firstly design target-like templet for object in forw... View full abstract»

Lane mark detection and tracking is essential for advanced driver assistance systems. We propose a computationally efficient lane mark detection and tracking method for expressway that can robustly and accurately detect lane marks in an image. A small size detection window scanner moving in the region of interest to determine whether there is a lane mark at the current position. This method can im... View full abstract»

As traditional Canny operator is manually determined in the threshold for edge detection, we presented an improved Canny algorithm based on adaptive two-dimensional Otsu and Newton Iterative. The algorithm bases on gray distribution characteristics of underwater suspended particles, it uses inter-class variance (Otsu) adaptive algorithm to determine the optimal threshold to connect and trace edge.... View full abstract»

Scene text detection is an important research problem in computer vision community. It has great application value in many fields. Inspired by Faster-RCNN which is a popular method for object detection, we consider to apply the Regional Proposal Network (RPN) method for scene text detection because text can be regarded as the common object. The core of RPN is to detect different sizes of objects w... View full abstract»

The fine-grained object detection is an extremely challenging problem due to the subtle variances in the appearances. At present, faster R-CNN is one of the best detection systems. However, it not a wise decision to directly apply the faster R-CNN to the fine-grained object detection. By analyzing the characteristics of fine-grained objects, we found that the anchor mechanism in the faster R-CNN s... View full abstract»

When beginners practice Chinese calligraphy, they often copy from ancient calligraphic works and try to imitate the style as closely as possible. However there are inevitably some characters whose styles are not correctly followed. Thus we are motivated to detect the style consistency of all written characters in one practice. With the styles extracted by using stacked autoencoders of deep neural ... View full abstract»

This paper studies an image descriptor that mimics the retina's photo receiving cell pattern. Various pattern differencing combinations and second order differencing techniques were explored. This new method shows higher precision and recall performance than the classical BRIEF and steered BRIEF descriptors. View full abstract»

In the medical image analysis, ROI (Region of Interest) is one of the key features of clinical diagnostic analysis. The applying of local features of ROI to the deep learning of image classification has the advantage of noise eliminating and information reducing. Based on existing research results, using Scale Invariant Feature Transformation (SIFT) algorithm combined with SVM classifier and slidi... View full abstract»

The higher-order statistics based independent component analysis (ICA) algorithm can extract natural image features. Based on the maximum likelihood ICA criterion, and using the weighted orthogonal constrained natural gradient, a new ICA algorithm is proposed. Natural image feature extraction simulation results show that, compared with other ICA algorithms, the proposed algorithm has faster conver... View full abstract»

This paper defines the SIFT-like features by analogy and proposes a novel method to accelerate its matching process. The acceleration strategy is to compute a characteristic value for each key point descriptor and divide a key point set into different subsets making use of this value. The approximate nearest neighbor (ANN) search method is applied to improve the efficiency of matching. The perform... View full abstract»