The most common impulse noise suppression methods are based on the median filter. They are suitable for low noise densities, but employing a separate noise detection step makes them suitable for higher noise rates. In...
详细信息
The most common impulse noise suppression methods are based on the median filter. They are suitable for low noise densities, but employing a separate noise detection step makes them suitable for higher noise rates. Instead, this may degrade the performance for lower noise. Particularly, the adaptive median noise removal algorithm improves the results for both low and high noise corruption rates. This paper presents a novel noise detection and filtering algorithm that uses leaky integrate and fire spiking neurons which model natural computing of the brain. In the proposed method, at each process of the detection and the filtering, all pixels of an adaptive sliding window are fed to the spiking neurons performing the operations in parallel. As a result, higher processing speed and higher operating frequency are achieved. An FPGA is used to implement the algorithm due to its real-time application and reconfigurable structure. The proposed architecture represents a slight increase in the PSNR values, whereas it consumes less hardware resources compared to the previous works. The design is implemented on Cyclone iv device from Altera family. It includes 3231 LUTs and the maximum operating frequency of 187 MHz is achieved.
作者:
Zhang, ZhengxiZhao, LiangLiu, YunanZhang, ShanshanYang, JianPCA Lab
Key Lab of Intelligent Perception and Systems for High -Dimensional Information of Ministry of Education Jiangsu Key Lab of Image and Video Understanding for Social Security School of Computer Science and Engineering Nanjing University of Science and Technology Nanjing China
It is an important yet challenging task to detect objects on hazy images in real-world applications. The major challenge comes from low visual quality and large haze density variations. In this work, we aim to jointly...
详细信息
Given the Covid-19 pandemic, the retail industry shifts many business models to enable more online purchases that produce large transaction data quantities (i.e., big data). Data science methods infer seasonal trends ...
详细信息
Considering the weaknesses of signature-based approaches adopted by current antimalware, from both academic and industrial side there is a boost in the development of techniques exploiting artificial intelligence, whe...
详细信息
Considering the weaknesses of signature-based approaches adopted by current antimalware, from both academic and industrial side there is a boost in the development of techniques exploiting artificial intelligence, where one of the most promising are based on the representation of application under analysis as image. In order to understand whether these approaches can be effectively adopted in the real-world, starting from a detector based on deep learning, in this paper we evaluate the resilience of these approaches when morphed samples are considered. We present DexWave, a tool aimed to automatically inject perturbations techniques targeting the smali code representation of Android applications. The experimental analysis demonstrate that image-based malware classifier are vulnerable to simple perturbations attack.
Domain Adaptation for semantic segmentation is of vital significance since it enables effective knowledge transfer from a labeled source domain (i.e., synthetic data) to an unlabeled target domain (i.e., real images),...
详细信息
Domain Adaptation for semantic segmentation is of vital significance since it enables effective knowledge transfer from a labeled source domain (i.e., synthetic data) to an unlabeled target domain (i.e., real images), where no effort is devoted to annotating target samples. Prior domain adaptation methods are mainly based on image-to-image translation model to minimize differences in image conditions between source and target domain. However, there is no guarantee that feature representations from different classes in the target domain can be well separated, resulting in poor discriminative representation. In this paper, we propose a unified learning pipeline, called image Translation and Representation Alignment (ITRA), for domain adaptation of segmentation. Specifically, it firstly aligns an image in the source domain with a reference image in the target domain using image style transfer technique (e.g., CycleGAN) and then a novel pixel-centroid triplet loss is designed to explicitly minimize the intra-class feature variance as well as maximize the inter-class feature margin. When the style transfer is finished by the former step, the latter one is easy to learn and further decreases the domain shift. Extensive experiments demonstrate that the proposed pipeline facilitates both image translation and representation alignment and significantly outperforms previous methods in both GTA5 → Cityscapes and SYNTHIA → Cityscapes scenarios.
Simultaneous localization and map construction (SLAM) tasks have been proven to benefit greatly from the depth information of the environment. In this paper, we first present an unsupervised end-to-end learning framew...
详细信息
Simultaneous localization and map construction (SLAM) tasks have been proven to benefit greatly from the depth information of the environment. In this paper, we first present an unsupervised end-to-end learning framework for the task of monocular depth and camera motion estimation from video sequences. The difference between our work and the existing unsupervised methods is that we not only use image reconstruction for supervising but also exploit the pose estimation method used in traditional SLAM approaches to enhance the supervised signal and add extra training constraints for the task of monocular depth and camera motion estimation. Furthermore, we successfully exploit our unsupervised learning framework to assist the traditional ORB-SLAM system when the initialization module of ORB-SLAM method could not match enough features. Qualitative and quantitative experiments have shown that our unsupervised learning framework performs the depth estimation task superior to the supervised methods and outperforms the previous state-of-the-art unsupervised approach by 13.5% on KITTI dataset. For the pose estimation task, our method performs comparably to the supervised methods that use ground-truth pose data for training. Besides, our unsupervised learning framework can significantly accelerate the initialization process of the traditional ORB-SLAM system and effectively improve the accuracy of environmental mapping in strong lighting and weak texture scenes.
This paper studies the traditional target classification and recognition algorithm based on Histogram of Oriented Gradients (HOG) feature extraction and Support Vector Machine (SVM) classification and applies this alg...
This paper studies the traditional target classification and recognition algorithm based on Histogram of Oriented Gradients (HOG) feature extraction and Support Vector Machine (SVM) classification and applies this algorithm to distributed artificial intelligence image recognition. Due to the huge number of images, the general detection speed cannot meet the requirements. We have improved the HOG feature extraction algorithm. Using principal component analysis (PCA) to perform dimensionality reduction operations on HOG features and doing distributed artificial intelligence image recognition experiments, the results show that the image detection efficiency is slightly improved, and the detection speed is also improved. This article analyzes the reason for these changes because PCA mainly uses the useful feature information in HOG features. The parallelization processing of HOG features on graphics processing unit (GPU) is studied. GPU is used for high parallel and high-density calculations, and the calculation of HOG features is very complicated. Using GPU for parallelization of HOG features can make the calculation speed of HOG features improved. We use image experiments for the parallelized HOG feature algorithm. Experimental simulations show that the speed of distributed artificial intelligence image recognition is greatly improved. By analyzing the existing digital image recognition methods, an improved BP neural network algorithm is proposed. Under the premise of ensuring accuracy, the recognition speed of digital images is accelerated, the time required for recognition is reduced, real-time performance is guaranteed, and the effectiveness of the algorithm is verified.
Low-level object matching can be done using projection signatures. In case of a large number of projections, the matching algorithm has to deal with less significant slices. A trivial approach would be to do statistic...
详细信息
Low-level object matching can be done using projection signatures. In case of a large number of projections, the matching algorithm has to deal with less significant slices. A trivial approach would be to do statistical analysis or apply machine learning to determine the significant features. To take adjacent values of the projection matrices into account, a convolutional neural network should be used. To compare two matrices, a Siamese structure of convolutional heads can be applied. In this paper, an experiment is designed and implemented to analyze the object matching performance of Siamese Convolutional Neural Networks based on multi-directional image projection data. A backtracking search-based Neural Architecture Generation method is used to create convolutional architectures, and a Master/Worker structured distributedprocessing with highly efficient scheduling based on the Longest processing Times-heuristics is used for parallel training and evaluation of the models. Results show that the projection-based methods are Pareto optimal in terms of one-shot classification accuracy and memory consumption.
The modern synchrotron radiation facilities are producing massive diffraction images, which present a severe problem for data processing due to the high dimensionality of imaging data. Feature recognition and selectio...
详细信息
The modern synchrotron radiation facilities are producing massive diffraction images, which present a severe problem for data processing due to the high dimensionality of imaging data. Feature recognition and selection based deep learning methods have been developed to analyze data automatically. One crucial step is to use AI to screen out the diffraction images without Bragg spots. This paper proposes a feature distillation based approach for screening. It helps to reduce over 40% raw data volume and greatly alleviates the post processing workload faced by scientists.
Edge detection algorithm has an important application in video imageprocessing. The algorithm overcomes the shortcomings of traditional video image edge detection methods, and significantly improves the quality of vi...
详细信息
暂无评论