imagethresholding is a useful method in many imageprocessing and computervision applications. However, it is not always satisfactory in all applications because of non-uniform illuminations. Otsu's method has b...
详细信息
ISBN:
(纸本)9780769528410
imagethresholding is a useful method in many imageprocessing and computervision applications. However, it is not always satisfactory in all applications because of non-uniform illuminations. Otsu's method has been widely used as the classical technique in real thresholding tasks. In this paper, we propose a novel method for adaptive local thresholding by applying Otsu's results. Based on simulated annealing, the proposed algorithm searches the optimal threshold for each partitioned subimage according to the quadtree data structure. It is also scale invariant for different object sizes. For reducing the computations, an improvement of Otsu's method is also developed. Experimental results show the efficiency of the proposed method.
Anomaly detection is a major task in crowd management through video surveillance. It refers to the events which are deviated from normal events. We have introduced an unsupervised method to detect motion anomaly in su...
详细信息
ISBN:
(纸本)9781450366151
Anomaly detection is a major task in crowd management through video surveillance. It refers to the events which are deviated from normal events. We have introduced an unsupervised method to detect motion anomaly in surveillance video. In this work we have considered only optical flow as feature. First, for each frame we compute magnitude of optical flow of motion using flownet2 [10]. then, mean of magnitude of flow due to regular normal motion (given in training data) is computed at each pixel where such motion exists in the training video frames. Our strategy is to compare the motion under consideration against this mean flow magnitude, and we expect that the anomalous motion would differ significantly from normal motion. An autoencoder type network is trained to detect this anomaly. Training data patch is constructed by Interleaving the columns of mean optical flow patch and the corresponding flow patch from each frame. this interleaving is done to incorporate context dependency. the autoencoder is trained to minimize mean-square reconstruction error between input column wise interleaved patch and output (i.e., reconstructed patch) of the autoencoder. During testing, a patch is declared anomalous if the reconstruction error is high compared to the training error. Experiments have been carried out on UCSD and UMN dataset and are compared with other methods. Our method gives comparable results with other state-of-the-art methods.
this book constitutes the refereed proceedings of the 4th International conference on Pattern Recognition and Machine Intelligence, PReMI 2011, held in Moscow, Russia in June/July 2011. the 65 revised papers presented...
ISBN:
(数字)9783642217869
ISBN:
(纸本)9783642217852
this book constitutes the refereed proceedings of the 4th International conference on Pattern Recognition and Machine Intelligence, PReMI 2011, held in Moscow, Russia in June/July 2011. the 65 revised papers presented together with 5 invited talks were carefully reviewed and selected from 140 submissions. the papers are organized in topical sections on pattern recognition and machine learning; image analysis; image and video information retrieval; natural language processing and text and data mining; watermarking, steganography and biometrics; soft computing and applications; clustering and network analysis; bio and chemo analysis; and document imageprocessing.
computer visibility is a multi-sectoral systematic field concerned with how computers assist to improve high-level understanding of digital images and videos. Haze is an atmospheric natural phenomenon that hinders com...
详细信息
ISBN:
(纸本)9781665434027
computer visibility is a multi-sectoral systematic field concerned with how computers assist to improve high-level understanding of digital images and videos. Haze is an atmospheric natural phenomenon that hinders computervision tasks. In the present era, image degradation is a major issue, and creating a high-quality, haze-free image remains a difficult challenge. As a consequence, the performance of many multimedia processing systems leads to poor object recognition and tracking. More haze removal algorithms have been proposed to counter this issue, which aesthetically enhances the visibility of visuals that have been deteriorated by adverse weather, hence eradicating the haze effect. Since haze is reliant on an uncertain scene depth, it is a significant challenge to resolve. this study investigated existing strategies for eradicating haze from obtained photographs. this review is designed to formulate strategies for researchers to point the appropriate direction for future advancements based on current accomplishments.
Underwater image quality is often compromised due to factors like fluorescence, low illumination, absorption, and scattering. Recent advancements in underwater image enhancement have introduced various deep network ar...
详细信息
ISBN:
(纸本)9798400710759
Underwater image quality is often compromised due to factors like fluorescence, low illumination, absorption, and scattering. Recent advancements in underwater image enhancement have introduced various deep network architectures to address these issues. Typically, these methods employ a single network to tackle all degradation challenges. However, we hypothesize that deep networks trained for specific conditions outperform those trained for multiple degradation cases. Consequently, we propose an iterative framework that individually identifies and resolves a dominant degradation condition. We focus on eight specific degradation conditions: low illumination, low contrast, haziness, blurriness, noise, and color imbalances across three channels. Our approach involves designing a deep network capable of detecting the dominant degradation condition and selecting an appropriate deep network tailored for that specific condition. We further propose that our work is the creation of condition-specific datasets derived from high-quality images in two standard datasets, UIEB and EUVP. these datasets facilitate the training of enhancement networks specific to each degradation condition. Our proposed method performs better than the nine baseline methods on both UIEB and EUVP datasets.
Defects on the Pipeline surface such as cracks cause main problems for governments, specifically when the pipeline is covered under the ground. Manual examination for surface defects in the pipeline has several disadv...
详细信息
Developing computervision applications is a challenging task. Realizing such an application as an embedded system is even more difficult, due to the limited resources of the target platform. In this paper the methods...
详细信息
ISBN:
(纸本)0889865981
Developing computervision applications is a challenging task. Realizing such an application as an embedded system is even more difficult, due to the limited resources of the target platform. In this paper the methods for the realization of such an embedded vision application are introduced. this includes a description of a workflow that supports the development of an embedded vision system. Also, the tools used for development and testing of this application are presented. For each step in the development process either off-the-shelf or self-developed tools were deployed. the embedded system, a stereo vision sensor, was part of an autonomous vehicle which participated in the DARPA Grand Challenge 2005. this system was used for obstacle and lane detection in this competition.
Person re-identification addresses the problem of matching pedestrian images across disjoint camera views. Design of feature descriptor and distance metric learning are the two fundamental tasks in person re-identific...
详细信息
ISBN:
(纸本)9781450366151
Person re-identification addresses the problem of matching pedestrian images across disjoint camera views. Design of feature descriptor and distance metric learning are the two fundamental tasks in person re-identification. In this paper, we propose a metric learning framework for person re-identification, where the discriminative metric space is learned using Kernel Fisher Discriminant Analysis (KFDA), to simultaneously maximize the inter-class variance as well as minimize the intra-class variance. We derive a Mahalanobis metric induced by KFDA and argue that KFDA is efficient to be applied for metric learning in person re-identification. We also show how the efficiency of KFDA in metric learning can be further enhanced for person re-identification by using two simple yet efficient multiple kernel learning methods. We conduct extensive experiments on three benchmark datasets for person re-identification and demonstrate that the proposed approaches have competitive performance with state-of-the-art methods.
Immuonohistochemically images of meningioma which are stained by ki67 marker contain positive and negative cells. Accurate counting the number of positive and negative cells in such images play a critical role in diag...
详细信息
Due to cardiovascular diseases being the main cause of death in Germany in 2020, about 298,557 stents, which can be used to treat these kinds of diseases, were implanted. Stents are medical devices and therefore, need...
详细信息
暂无评论