IEEE Transactions on Pattern Analysis and Machine Intelligence

Video compression algorithms have been designed aiming at pleasing human viewers, and are driven by video quality metrics that are designed to account for the capabilities of the human visual system. However, thanks to the advances in computer vision systems more and more videos are going to be watched by algorithms, e.g. implementing video surveillance systems or performing automatic video tagging. This paper describes an adaptive video coding approach for computer vision-based systems. We show how to control the quality of video compression so that automatic object detectors can still process the resulting video, improving their detection performance, by preserving the elements of the scene that are more likely to contain meaningful content. Our approach is based on computation of saliency maps exploiting a fast objectness measure. The computational efficiency of this approach makes it usable in a real-time video coding pipeline. Experiments show that our technique outperforms standard H.265 in speed and coding efficiency, and can be applied to different types of video domains, from surveillance to web videos.

/pdf/video-compression-for-object-detection-algorithms-4v5cejebar.pdf

Video Compression for Object Detection Algorithms

Various external environments can affect detection for insulator from the transmission line in aerial image. Segmenting the insulator from the image with the traditional image segmentation algorithm will influence the effect of detection and recognition. An image segmentation method for insulator based on ant colony classification and weighted variable fuzzy c-means (FCM) was proposed in this study to overcome the limitations of the traditional detection method for insulator in transmission line. An image enhancement method based on improved fuzzy set was introduced in the image enhancement stage. The sine membership function was used to enhance the preprocessed image according to the gray feature of insulator image. An image segmentation method based on weighted variable FCM was designed in the image segmentation stage. Different weights were defined to represent the difference between neighborhood pixels and their central pixel to divide the insulator from the transmission line. Experimental results demonstrate that the error rate of the proposed algorithm is less than 0.02 when three group insulator images (50 images per group) are randomly extracted and the signal-to-noise ratio (SNR) of the image is more than 30. The average time of image segmentation is 4.96 times lower than that of the traditional FCM. These findings indicate that the proposed algorithm fully utilizes the spatial features of the image and improves the effect and efficiency of image segmentation for insulator. The proposed method provides technical support for subsequent recognition and defect diagnosis of insulator.

/pdf/image-segmentation-method-of-insulator-in-transmission-line-3entsphlu7.pdf

Image Segmentation Method of Insulator in Transmission Line Based on Weighted Variable Fuzzy C-Means

Moving target detection (MTD) is one of the emphases and difficulties in the field of computer vision and image processing. It is the basis of moving target tracking and behavior recognition. We propose two methods are improved and fused, respectively, and the fusion algorithm is applied to the complex scene for MTD, so as to improve the accuracy of MTD in complex and hybrid scenes. Using the main idea of the three-frame difference image method, the background difference method and the interframe difference method are combined to make their advantages complementary to overcome each other’s weaknesses. The experimental results show that the method can be well adapted to the situation of periodic motion interference in the background, and it can adapt to the situation of sudden background changes.

/pdf/design-of-sports-image-contour-feature-acquisition-system-1gf37zyt.pdf

Design of Sports Image Contour Feature Acquisition System Based on the Background Subtraction Method

The traditional saliency detection obtains significant regions by using the region contrast of color information. In complex and multi target images, the method often performs poorly, and the significant region can not be detected completely. Therefore, a significant region detection method based on accurate edge learning is proposed in this paper. First, in order to achieve better image cutting effect, we use multiple WMF (weighted median filter) to process the source image and then use the method of graph cut to divide the image to get some regions with similar features, that is, the segmentation graph. At the same time, we use background prior method to get the preliminary salient map. The edge learning area is obtained by combining the initial saliency map and the bootstrap map. The area is the foreground and the other regions are the background, and the convolution neural network VGG16 model is used to complete the detection. The algorithm is verified on the open dataset ECSSD. The results show that the algorithm has better recall and precision.

Significant Regional Detection Based on Precise Edge Learning

This study presents a visual attention model for determining image areas to receive different extents of video compression in order to minimize perceived program artefacts whilst maximizing the compression possible. The model integrates features related to motion with existing video image compression algorithms. The proposed visual attention model extracts the color, intensity, textural, and motion features of a video to determine the predicted region of interest (ROI). First, color, intensity, and texture saliency maps are generated by applying the "center-surround" method and a motion saliency map is produced using a difference operator. Then, a multi-channel weighting method is used to generate a global saliency map and to determine the ROI according to a winner-takes-all network (WTA). The proposed video image compression algorithm performs either low or no compression on the ROI while a high degree of compression is applied to the other regions. Tests indicate that the proposed visual attention model is able swiftly to identify the ROI, allowing the proposed compression algorithm to exert a high compression efficiency yet with minimally noticeable visual degradation.

http://dline.info/fpaper/jdim/v12i5/7.pdf

Hongchang Ke

Papers

A Video Image Compression Method based on Visually Salient Features