Fast Image Dehazing Method Based on Linear Transformation

Fast Image Dehazing Method Based on Linear Transformation

ABSTRACT:

 Images captured in hazy or foggy weather conditions are seriously degraded by the scattering of atmospheric particles, which directly influences the performance of outdoor computer vision systems. In this paper, a fast algorithm for single image dehazing is proposed based on linear transformation by assuming that a linear relationship exists in the minimum channel between the hazy image and the haze-free image. First, the principle of linear transformation is analyzed. Accordingly, the method of estimating a medium transmission map is detailed and the weakening strategies are introduced to solve the problem of the brightest areas of distortion. To accurately estimate the atmospheric light, an additional channel method is proposed based on quad-tree subdivision. In this method, average grays and gradients in the region are employed as assessment criteria. Finally, the haze-free image is obtained using the atmospheric scattering model. Numerous experimental results show that this algorithm can clearly and naturally recover the image, especially at the edges of sudden changes in the depth of field. It can thus achieve a good effect for single image dehazing. Furthermore, the algorithmic time complexity is a linear function of the image size. This has obvious advantages in running time by guaranteeing a balance between the running speed and the processing effect.

 EXISTING SYSTEM:

  • They proposed a simple but powerful color attenuation prior for haze removal from a single input hazy image.
  • By creating a linear model for modeling the scene depth of the hazy image under this novel prior and learning the parameters of the model with a supervised learning method, the depth information can be well recovered.
  • With the depth map of the hazy image, they can easily estimate the transmission and restore the scene radiance via the atmospheric scattering model, and thus effectively remove the haze from a single image.

 DISADVANTAGES OF EXISTING SYSTEM:

  • Existing dehazing methods can achieve good results on a single image, the high computation limits their applications in real-time systems.

 PROPOSED SYSTEM:

  • In this method, the transmission map is estimated using a linear transformation model which has less computational complexity, and the atmospheric light is obtained with an additional channel method based on a quad-tree subdivision by using the ratio of grays and gradients in the region.
  • With those information, we can easily get the haze-free image through the atmospheric scattering model.

ADVANTAGES OF PROPOSED SYSTEM:

  • In this method, the estimation of the transmission map is based on a linear model that includes only linear operations without any exponential operations or sample training. Therefore, it is easy to realize and has less computational complexity.

 SYSTEM ARCHITECTURE:

 SYSTEM REQUIREMENTS:

HARDWARE REQUIREMENTS:

  • System : Pentium IV 2.4 GHz.
  • Hard Disk : 40 GB.
  • Floppy Drive : 44 Mb.
  • Monitor : 15 VGA Colour.
  • Mouse :
  • Ram : 512 Mb.

SOFTWARE REQUIREMENTS: 

  • Operating system : Windows XP/7.
  • Coding Language : MATLAB
  • Tool : MATLAB R2013A

REFERENCE:

Wencheng Wang, Xiaohui Yuan, Member, IEEE, Xiaojin Wu and Yunlong Liu, “Fast Image Dehazing Method Based on Linear Transformation”, IEEE Transactions on Multimedia, 2017.

Enhancing Automatic Maritime Surveillance aSystems with Visual Information

Enhancing Automatic Maritime Surveillance aSystems with Visual Information

 ABSTRACT:

Automatic surveillance systems for the maritime domain are becoming more and more important due to a constant increase of naval traffic and to the simultaneous reduction of crews on decks. However, available technology still provides only a limited support to this kind of applications. In this paper, a modular system for intelligent maritime surveillance, capable of fusing information from heterogeneous sources, is described. The system is designed to enhance the functions of the existing vessel traffic services systems and to be deployable in populated areas, where radar-based systems cannot be used due to the high electromagnetic radiation emissions. A quantitative evaluation of the proposed approach has been carried out on a large and publicly available data set of images and videos, which are collected from multiple real sites, with different light, weather, and traffic conditions.

 EXISTING SYSTEM:

  • An open problem in the maritime scenario is how to fuse information from the different sensors.
  • Bustamante et al. paper proposed a multi-agent system (MAS) architecture for automatically controlling the camera, radar, and AIS modules.
  • The data fusion system is represented as an additional source sensor, which allows the agents to collaborate for avoiding redundancy.

 DISADVANTAGES OF EXISTING SYSTEM:

  • Existing system is used for avoiding redundancy. However, the conflict solving mechanism between agents can cause lost time.

 PROPOSED SYSTEM:

  • In this paper, we focus on Vessel Traffic Services (VTS) systems, which combine radar and AIS data and are often equipped with long-range surveillance cameras, both electro-optical (EO) and infra-red (IR). However, using radar and AIS data only is not sufficient to ensure a complete solution for the maritime surveillance problem, due to two strong limitations:

1) AIS signal may be not available (AIS device not activated or malfunctioning) or illegally manipulated.

2) Radar-based systems are not suitable for vessel traffic monitoring in populated areas, due to the high electromagnetic radiation emissions.

  • Replacing radar sensors with cameras is a feasible solution for the maritime surveillance task, without the need of placing radar antennas in populated areas. Here, we propose a modular architecture that extends the capability of currently available VTS systems, together with a prototype system that allows adds a novel visual dimension to the common VTS features. The architecture is designed for:

1) Detecting boats through a classifier-based method, which can work with both EO and IR moving cameras.

2) Tracking multiple ships, even in presence of occlusions.

3) Fusing data from existing VTS systems with visual information from cameras.

4) Deployable in populated areas.

 ADVANTAGES OF PROPOSED SYSTEM:

  • A major advantage of the proposed approach is the possibility of providing a global view of the captured scene, by adding a visual dimension to radar and AIS data, which is very effective for the user.

SYSTEM ARCHITECTURE:

 SYSTEM REQUIREMENTS:

HARDWARE REQUIREMENTS: 

  • System : Pentium IV 2.4 GHz.
  • Hard Disk : 40 GB.
  • Floppy Drive : 44 Mb.
  • Monitor : 15 VGA Colour.
  • Mouse :
  • Ram : 512 Mb.

SOFTWARE REQUIREMENTS: 

  • Operating system : Windows XP/7.
  • Coding Language : MATLAB
  • Tool : MATLAB R2013A

REFERENCE:

Domenico D. Bloisi, Fabio Previtali, Andrea Pennisi, Daniele Nardi, and Michele Fiorini, Senior Member, IEEE, “Enhancing Automatic Maritime Surveillance Systems With Visual Information”, IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS, 2017.

Discriminative Multi-view Interactive Image Re-ranking

Discriminative Multi-view Interactive Image Re-ranking

ABSTRACT:

Given unreliable visual patterns and insufficient query information, content-based image retrieval (CBIR) is often suboptimal and requires image re-ranking using auxiliary information. In this paper, we propose Discriminative Multi-view Interactive Image Re-ranking (DMINTIR), which integrates User Relevance Feedback (URF) capturing users’ intentions and multiple features that sufficiently describe the images. In DMINTIR, heterogeneous property features are incorporated in the multi-view learning scheme to exploit their complementarities. In addition, a discriminatively learned weight vector is obtained to reassign updated scores and target images for reranking. Compared to other multi-view learning techniques, our scheme not only generates a compact representation in the latent space from the redundant multi-view features but also maximally preserves the discriminative information in feature encoding by the large-margin principle. Furthermore, the generalization error bound of the proposed algorithm is theoretically analyzed and shown to be improved by the interactions between the latent space and discriminant function learning. Experimental results on two benchmark datasets demonstrate that our approach boosts baseline retrieval quality and is competitive with other state-of-the-art re-ranking strategies.

EXISTING SYSTEM:

  • In existing paper, they adopt click data to bridge the semantic gap. They proposed a novel multi-view hypergraph-based learning (MHL) method that adaptively integrates click data with varied visual features.
  • In particular, MHL considers pairwise discriminative constraints from click data to maximally distinguish images with high click counts from images with no click counts, and a semantic manifold is constructed. It then adopts hypergraph learning to build multiple manifolds from varied visual features.
  • Finally, MHL integrates the semantic manifold with visual manifolds through an iterative optimization procedure. The weights of different manifolds and the re-ranking score are simultaneously obtained after using this optimization strategy.

DISADVANTAGES OF EXISTING SYSTEM:

Multi-view re-ranking methods have shown some promise, they still suffer from one or more of the following limitations.

  • First, a compact representation that maintains sufficient discriminative power needs to be efficiently produced from multi-view feature spaces, while considering the demand on retrieval efficiency and accuracy.
  • Second, the features used for fusion should be carefully chosen. Most existing multi-view image re-ranking approaches focus on ad-hoc low-level image representations (e.g., color histograms), with the resulting relatively poor initial retrieval accuracy limiting further performance improvements by re-ranking.
  • Third, user interactive information can be used as invaluable supervised information for multi-view image reranking, whereas it is usually downplayed in standard practice. Finally, thorough theoretical analyses are indispensable for interpreting algorithm properties but are not always available.

 PROPOSED SYSTEM:

Our method can be divided into two components, namely training and testing. In model training, we aim to build a re-ranking model. Specifically, given the initial ranking list, we first exploit interactive user relevance feedback to label some positive images as well as low-ranked negative images obtained by pseudo relevance feedback. These training examples are represented using multi-view features, which are then integrated into a latent subspace via our proposed algorithm. Meanwhile, discriminative embedding is performed in this subspace such that a discriminative boundary is derived and the latent representations sufficiently preserve the separating capability. During the test phase, we will obtain the latent representations of all images in the database based on the learned model. Then, we re-evaluate the relevance of all images w.r.t. the query by their signed distances from the discriminative boundary. Since our proposed algorithm is for re-ranking, it is independent of the ranking method generating the initial ranking results.

The contributions of this paper are summarized as follows:

  1. We exploit the complementarity between the deep features and shallow representations for our multi-view learning. To our knowledge, this is first attempt at integrating these two heterogeneous features into a multiview feature learning model.
  2. We propose an accurate interactive re-ranking model coined DMINTIR. By leveraging user relevance feedback (URF), we derive a robust representation by projecting the original multiple features onto a latent subspace while the discriminative information is maximally preserved. Thus, the multi-view feature learning and discriminative embedding are simultaneously unified into a generic formulation.
  3. We explore the generalization error bound of the proposed algorithm and demonstrate the benefits of combining multi-view feature learning with discriminative embedding in our theoretical analysis.

 ADVANTAGES OF PROPOSED SYSTEM:

  • To overcome existing limitations, we proposed a discriminative approach for multi-view interactive image re-ranking called DMINTIR. We derived the latent representation to losslessly recover the original feature space.
  • The original and latent spaces can be connected via a generation function rather than by computationally expensive graph embedding. We unify current state-of-the-art CNN codes and shallow architecture features built on local descriptors into multi-view learning for robust feature encoding.
  • Meanwhile, User Relevance Feedback (URF) is employed for interactive re-ranking, and a discriminative model is obtained by large margin optimization.
  • Thus, we obtain a generic and unified framework for accurate image re-ranking. Furthermore, the generalization error bound of the proposed algorithm is theoretically analyzed.

SYSTEM ARCHITECTURE:

 SYSTEM REQUIREMENTS:

HARDWARE REQUIREMENTS: 

  • System : Pentium IV 2.4 GHz.
  • Hard Disk : 40 GB.
  • Floppy Drive : 44 Mb.
  • Monitor : 15 VGA Colour.
  • Mouse :
  • Ram : 512 Mb.

SOFTWARE REQUIREMENTS:

  • Operating system : Windows XP/7.
  • Coding Language : MATLAB
  • Tool : MATLAB R2013A

REFERENCE:

Jun Li, Chang Xu, Wankou Yang, Changyin Sun and Dacheng Tao, Fellow, IEEE, “Discriminative Multi-view Interactive Image Re-ranking”, IEEE Transactions on Image Processing, 2017.

Counting vehicles in urban traffic scenes using foreground time-spatial images

Counting vehicles in urban traffic scenes using foreground time-spatial images

ABSTRACT:

A foreground time-spatial image (FTSI) is proposed for counting vehicles in complex urban traffic scenes to resolve deficiencies of traditional counting methods, which are highly computationally expensive and become unsuccessful with increasing complexity in urban traffic scenarios. First, a self-adaptive sample consensus background model with confidence measurements for each pixel is constructed on the virtual detection line in the frames of a video. The foreground of the virtual detection line is then collected over time to form a FTSI. The occlusion cases are then estimated based on the convexity of connected components. Finally, counting the number of connected components in the FTSI reveals the number of vehicles. Based on real-world urban traffic videos, the experiments in this study are conducted using FTSI, and compared in accuracy with two other time-spatial images methods. Experimental results based on real-world urban traffic videos show that the accuracy rate of the proposed approach is above 90% and it performs better than the state-of the-art methods.

EXISTING SYSTEM:

  • In exiting they proposed a real-time cost-effective traffic monitoring system that can reliably perform traffic flow estimation and vehicle classification at the same time.
  • First, the foreground is extracted using a pixel-wise weighting list that models the dynamic background. Shadows are discriminated utilizing color and edge invariants.
  • Second, the foreground on a specified check-line is then collected over time to form a spatial-temporal profile image.
  • Third, the traffic flow is estimated by counting the number of connected components in the profile image.
  • Finally, the vehicle type is classified according to the size of the foreground mask region.

 DISADVANTAGES OF EXISTING SYSTEM:

  • The TSI in existing is built generating the foreground vehicle based on background subtraction, and BS-based vehicle detection methods and TSI are combined for counting vehicle. However, this method cannot deal with slow-moving or temporarily stopped vehicles.

PROPOSED SYSTEM:

  • In this paper, a time-spatial foreground image is proposed for vehicle counting in complex urban traffic scenes to resolve deficiencies in traditional counting methods that are computationally expensive and failure-prone in complex urban traffic scenarios.
  • A self-adaptive sample consensus background model with confidence measurement at each pixel location is constructed only on the virtual detection line of the video frames. A time-spatial foreground image is stacked over time based on the foreground determined by the virtual detection line.
  • After occlusion cases are estimated based on the convexity of connected components obtained, the number of connected components in the FTSI is counted to obtain the number of vehicles.

 ADVANTAGES OF PROPOSED SYSTEM:

  • The detection accuracy analyses performed demonstrated that the proposed method achieves better performance by both qualitative and quantitative measures compared with other TSI methods suggested in the literature.
  • Experimental results using real-world urban traffic videos show that the accuracy the of proposed approach is above 90% and that it performs better than other state-of-the-art methods.

 SYSTEM REQUIREMENTS:

HARDWARE REQUIREMENTS:

  • System : Pentium IV 2.4 GHz.
  • Hard Disk : 40 GB.
  • Floppy Drive : 44 Mb.
  • Monitor : 15 VGA Colour.
  • Mouse :
  • Ram : 512 Mb.

SOFTWARE REQUIREMENTS:

  • Operating system : Windows XP/7.
  • Coding Language : MATLAB
  • Tool : MATLAB R2013A

REFERENCE:

Yunsheng Zhang1, Chihang Zhao1 , Qiuge Zhang, “Counting vehicles in urban traffic scenes using foreground time-spatial images”, IEEE 2017.

An Emotion Recognition System For Mobile Applications

An Emotion Recognition System For Mobile Applications

ABSTRACT:

Emotion-aware mobile applications have been increasing due to their smart features and user acceptability. To realize such an application, an emotion recognition system should be in real time and highly accurate. As a mobile device has limited processing power, the algorithm in the emotion recognition system should be implemented using less computation. In this paper, we propose an emotion recognition with high performance for mobile applications. In the proposed system, facial video is captured by an embedded camera of a smart phone. Some representative frames are extracted from the video, and a face detection module is applied to extract the face regions in the frames. The Bandlet transform is realized on the face regions, and the resultant subband is divided into non-overlapping blocks. Local binary patterns’ histograms are calculated for each block, and then are concatenated over all the blocks. The Kruskal-Wallis feature selection is applied to select the most dominant bins of the concatenated histograms. The dominant bins are then fed into a Gaussian mixture model based classifier to classify the emotion. Experimental results show that the proposed system achieves high recognition accuracy in a reasonable time.

EXISTING SYSTEM:

  • They introduced the architecture of an emotion-aware ambient intelligent and gerontechnological project named “Improvement of the Elderly Quality of Life and Care through Smart Emotion Regulation”.
  • The objective of the work is to find solutions for improving the quality of life and care of the elderly who can or want to continue living at home by using emotion regulation techniques.
  • A series of sensors is used for monitoring the elderlies’ facial and gestural expression, activity and behaviour, as well as relevant physiological data.
  • This way the older people’s emotions are inferred and recognized. Music, color and light are the stimulating means to regulate their emotions towards a positive and pleasant mood.

DISADVANTAGES OF EXISTING SYSTEM:

  • All these emotion recognition systems either used huge training data, or are computationally expensive.

PROPOSED SYSTEM:

In this paper, we propose a high-performance emotion recognition system for mobile applications. The embedded camera of a smart phone captures video of the user. The Bandlet transform is applied to some selective frames, which are extracted from the video, to give some subband images. Local binary patterns (LBP) histogram is calculated from the subband images. This histogram describes the features of the frames. A Gaussian mixture model (GMM) based classifier is used as a classifier. The proposed emotion recognition system is evaluated using several databases. The contribution of this paper is as follows:

  • the use of the Bandlet transform in emotion recognition,
  • the use of the Kruskal-Wallis (KW) feature selection to reduce the time requirement during a test phase, and
  • an achievement of higher accuracies in two publicly available databases compared with those using other contemporary systems.

ADVANTAGES OF PROPOSED SYSTEM:

  • The rate of accuracy is high.
  • It takes less than 1.4 seconds to recognize one instance of emotion. The high performance and the less time requirement of the system make it suitable to any emotion-aware mobile applications

SYSTEM ARCHITECTURE:

SYSTEM REQUIREMENTS:

HARDWARE REQUIREMENTS: 

  • System : Pentium IV 2.4 GHz.
  • Hard Disk : 40 GB.
  • Floppy Drive : 44 Mb.
  • Monitor : 15 VGA Colour.
  • Mouse :
  • Ram : 512 Mb.

SOFTWARE REQUIREMENTS:

  • Operating system : Windows XP/7.
  • Coding Language : MATLAB
  • Tool : MATLAB R2013A

REFERENCE:

SHAMIM HOSSAIN1, (Senior Member, IEEE), and GHULAM MUHAMMAD, “An emotion recognition system for mobile applications”, IEEE ACCESS, 2017.

 

A Robust and Efficient Approach to License Plate Detection

A Robust and Efficient Approach to License Plate Detection

ABSTRACT:

This paper presents a robust and efficient method for license plate detection with the purpose of accurately localizing vehicle license plates from complex scenes in real time. A simple yet effective image downscaling method is first proposed to substantially accelerate license plate localization without sacrificing detection performance compared with that achieved using the original image. Furthermore, a novel line density filter approach is proposed to extract candidate regions, thereby significantly reducing the area to be analyzed for license plate localization. Moreover, a cascaded license plate classifier based on linear SVMs using color saliency features is introduced to identify the true license plate from among the candidate regions. For performance evaluation, a dataset consisting of 3828 images captured from diverse scenes under different conditions is also presented. Extensive experiments on the widely used Caltech license plate dataset and our newly introduced dataset demonstrate that the proposed approach substantially outperforms state-of the-art methods in terms of both detection accuracy and run-time efficiency, increasing the detection ratio from 91.09% to 96.62% while decreasing the run time from 672 ms to 42 ms for processing an image with a resolution of 1082*728. The executable code and our collected dataset are publicly available.

PROJECT OUTPUT VIDEO: (Click the below link to see the project output video):

EXISTING SYSTEM:

  • They proposed a novel scheme to automatically locate license plates by principal visual word (PVW), discovery and local feature matching.
  • Observing that characters in different license plates are duplicates of each other, we bring in the idea of using the bag-of-words (BoW) model popularly applied in partial-duplicate image search.
  • Unlike the classic BoW model, for each plate character, we automatically discover the PVW characterized with geometric context.
  • Given a new image, the license plates are extracted by matching local features with PVW.

DISADVANTAGES OF EXISTING SYSTEM:

  • Most of the previous methods perform well only under certain predefined conditions.
  • Some common restrictions include fixed illumination, license plates with little blur or distortion from viewpoint changes, relatively simple backgrounds and the presence of only a single license plate in an image.
  • More recent state-of-the-art approaches impose fewer restrictions on license plate detection at the cost of increased computational complexity.
  • However, these approaches still have difficulty extracting license plates from complex scenes.

PROPOSED SYSTEM:

This paper develops an efficient and robust approach to license plate detection that is able to accurately localize one or multiple vehicle license plate(s) with diverse variations from complex backgrounds in real time. To speed up the detection algorithm overall, we first investigate how to reduce the size of the original high resolution image without decreasing license plate detection performance. Note that because of the negative effects that are generally introduced by the downsampling method that is commonly used in image processing, most previously developed methods perform license plate detection using the original image. Then, we analyze the common characteristics among diverse license plates and their major differences with respect to background regions to serve as a basis for designing a region filter to exclude irrelevant regions in the image. Furthermore, we study which features are most discriminative for license plate detection and then propose an efficient and robust classifier to ultimately localize the exact position of the license plate in the image.

In summary, the contributions of this paper are as follows:

1) A novel line density filter (LDF) is proposed to extract candidate license plate regions, thereby significantly reducing the area to be analyzed for license plate localization.

2) An efficient license plate verification method is proposed to accurately detect the true license plate from among the candidate regions using a cascaded license plate classifier (CLPC), which is trained based on color saliency features.

3) For performance evaluation, we present a newly collected challenging dataset that consists of 3828 license plate images with variations in illumination, license plate appearance, vehicle location and weather conditions.

4) We demonstrate that the proposed approach outperforms state-of-the-art methods by a large margin in terms of both detection accuracy and run-time efficiency.

ADVANTAGES OF PROPOSED SYSTEM:

  • Our approach is the fastest, requiring only 42 ms for license plate detection.
  • Our proposed method increasing the detection ratio from 91.09% to 96.62% while decreasing the run time from 672 ms to 42 ms for the processing of an image with a resolution of 1082×

SYSTEM ARCHITECTURE:

SYSTEM REQUIREMENTS:

HARDWARE REQUIREMENTS: 

  • System : Pentium IV 2.4 GHz.
  • Hard Disk : 40 GB.
  • Floppy Drive : 44 Mb.
  • Monitor : 15 VGA Colour.
  • Mouse :
  • Ram : 512 Mb.

SOFTWARE REQUIREMENTS: 

  • Operating system : Windows XP/7.
  • Coding Language : MATLAB
  • Tool : MATLAB R2013A

REFERENCE:

Yule Yuan, Member, IEEE, Wenbin Zou, Yong Zhao, Xinan Wang, Xuefeng Hu, and Nikos Komodakis, “A Robust and Efficient Approach to License Plate Detection”, IEEE TRANSACTIONS ON IMAGE PROCESSING, 2017.