The proceedings contain 90 papers. The topics discussed include: a GPU based real-time CUDA implementation for obtaining visual saliency;fingerprint enhancement using unsupervised hierarchical feature learning;a Schr&...
ISBN:
(纸本)1595930361
The proceedings contain 90 papers. The topics discussed include: a GPU based real-time CUDA implementation for obtaining visual saliency;fingerprint enhancement using unsupervised hierarchical feature learning;a Schrödinger formalism for simultaneously computing the Euclidean distance transform and its gradient density;duplication detection for image sharing systems;finding group interactions in social gathering videos;particle coding for meshfree cutting of deformable assets;total cluster: a person agnostic clustering method for broadcast videos;multispectral multifocus image fusion with guided steerable frequency and improved saliency;generalized synthesis and analysis prior algorithms with application to impulse denoising;and discovery of sparse contextual relationships for activity recognition in continuous videos.
This paper investigates compression of depth images-particularly, noisy depth images-captured by depth sensors like Kinect. Our scheme is based on incrementally detecting planes in depth images and then storing the pl...
详细信息
ISBN:
(纸本)1595930361
This paper investigates compression of depth images-particularly, noisy depth images-captured by depth sensors like Kinect. Our scheme is based on incrementally detecting planes in depth images and then storing the plane parameters instead of the depth information of the pixels lying on the detected planes. Residuals are then computed and compressed using standard image compression techniques. Our technique incorporates the input error model for comprehensive and accurate plane detection. Thereby, this accounts for the reliability of the input data in the compression scheme. The plane detection also accounts for edges. Experiments exhibit better image quality than standard compression techniques with smaller error. We additionally propose a novel error metric to evaluate compression of noisy depth images. Copyright is held by the authors.
Jump Flooding is a method for propagating labels across a given plane from different seeds. It has been used to compute the discrete Voronoi tessellation of a given plane efficiently. We introduce a version of JFA, wh...
详细信息
ISBN:
(纸本)1595930361
Jump Flooding is a method for propagating labels across a given plane from different seeds. It has been used to compute the discrete Voronoi tessellation of a given plane efficiently. We introduce a version of JFA, which optimizes the number of pixels processed by computing only the faces of the Voronoi tessellation. The pixels in the interior of the Voronoi regions are not processed resulting in a 1-skeleton representation of the Voronoi tessellation in 2D and a 2-skeleton representation in 3D. We describe an implementation of this algorithm on a GPU using CUDA and demonstrate its performance benefits on multiple data sets. As an application of the proposed algorithm, we present a GPU based method for extraction of channel centerlines in biomolecules. The fast computation of the discrete Voronoi diagram is exploited to extract channels in molecular dynamics simulation trajectories on-the-fly, thereby supporting the interactive visual analysis of static and dynamic channel structures. Copyright is held by the authors.
Given a video, there are many algorithms to separate static and dynamic objects present in the scene. The proposed work is focused on classifying the dynamic objects further as having either repetitive or non-repetiti...
详细信息
ISBN:
(纸本)9789897581335
Given a video, there are many algorithms to separate static and dynamic objects present in the scene. The proposed work is focused on classifying the dynamic objects further as having either repetitive or non-repetitive motion. In this work, we propose a novel approach to achieve this challenging task by processing the optical flow fields corresponding to the video frames of a dynamic natural scene. We design an unsupervised learning algorithm which uses functions of the flow vectors to design the feature vector. The proposed algorithm is shown to be effective in classifying a scene into static, repetitive, and non-repetitive regions. The proposed approach finds significance in various vision and computational photography tasks such as video editing, video synopsis, and motion magnification.
Color, texture, and shape are the low level features required for an effective image retrieval system. In this paper, we propose a hybrid approach which performs the fusion of color and texture features for image retr...
详细信息
ISBN:
(纸本)1595930361
Color, texture, and shape are the low level features required for an effective image retrieval system. In this paper, we propose a hybrid approach which performs the fusion of color and texture features for image retrieval. The color features are extracted by using the well known color histogram technique, while the texture features are captured using recently proposed features, known as orthogonal combination of linear binary pattern (OCLBP). The performance of the proposed approach is compared with a recently developed method, color difference histogram (CDH). Performance evaluation through extensive experiments reveals the superiority of the proposed approach over the state-of-the-art technique.
This paper proposes a real-time solution to setting up a virtual trial-room for on-line portals selling apparels using a generic web camera interface to the portal. The user selects an image of an apparel from the on-...
详细信息
ISBN:
(纸本)1595930361
This paper proposes a real-time solution to setting up a virtual trial-room for on-line portals selling apparels using a generic web camera interface to the portal. The user selects an image of an apparel from the on-line display and captures his/her own videos. The proposed method detects the pose of the user as well as various anthropomorphic features such as length and thickness of upper limbs and the dimensions of the torso. We use a background subtraction based methodology to segment out the human body from the image. The segmented human body contour is represented by a 1D curve by computing the distance of a point on the contour from the body centroid. Various extremities of body parts are found out by measuring the curvature. Using the detected feature points, we use a cloth fitting algorithm to fit the garment to the users body. The entire process is performed at 30fps, providing a realistic rendering of virtual clothing for any user.
In this paper we present a novel method for compression of focused plenoptic (light field) images. Plenoptic image is an array of lenslet images where the consecutive images act as stereo images. Adjacent lenslet imag...
详细信息
ISBN:
(纸本)1595930361
In this paper we present a novel method for compression of focused plenoptic (light field) images. Plenoptic image is an array of lenslet images where the consecutive images act as stereo images. Adjacent lenslet images have overlapping regions leading to higher levels of redundancy than observed in a normal image. We use this overlap to appropriately select a set of reference images from which the remaining lenslet images can be approximately reconstructed. These reference images can be compressed using jpeg. The residue of the reconstructed image from the original image is compressed using either DWT-SPIHT or DCT based compression. Using the proposed method we are able to compress the image at high PSNR at a low bits per pixel (bpp). Copyright 2014 ACM.
Camouaging an object in a photograph is normally performed with the intent of unnoticeably hiding it within a given image. In this work, we give a different dimension to this problem and raise the interesting issue of...
详细信息
ISBN:
(纸本)1595930361
Camouaging an object in a photograph is normally performed with the intent of unnoticeably hiding it within a given image. In this work, we give a different dimension to this problem and raise the interesting issue of camouaging motion blur with special relevance to non-uniformly blurred images. Given a blurred photograph, we apply a suitably derived blurring model to smear a target object and naturally blend it into the motion blurred background by using alpha matting. We validate our photo-realistic compositing approach on several synthetic and real examples. Copyright 2014 ACM.
Detection of text embedded in images and videos is very useful in applications like indexing and retrieval of multimedia content. There exist very few techniques which can differentiate between overlay text which is e...
详细信息
ISBN:
(纸本)1595930361
Detection of text embedded in images and videos is very useful in applications like indexing and retrieval of multimedia content. There exist very few techniques which can differentiate between overlay text which is embedded artificially and scene text which occurs naturally in images. This paper proposes an efficient method for detection and localization of overlay text from still images. By using colour filter array(CFA)-based forgery localization method, the proposed technique successfully removes any scene text present in the image. Experimental results on a database of 315 images demonstrate the efficiency of the proposed method. Copyright 2014 ACM.
Synthetic Aperture Radar (SAR) images are efficient and reliable source of information in extraction of damaged regions in case of floods. In assessment of damage accurately due to floods, image registration of optica...
详细信息
ISBN:
(纸本)1595930361
Synthetic Aperture Radar (SAR) images are efficient and reliable source of information in extraction of damaged regions in case of floods. In assessment of damage accurately due to floods, image registration of optical (before-flood) and SAR images (after-flood) has to be carried out efficiently. In this paper, we discuss a robust multi-sensor image registration algorithm using scale invariant feature points for keypoint extraction. For matching the keypoints, a multi-objective genetic algorithm is developed with angle, distance and vicinity criterion as the fitness functions. This optimization process helps in matching the scale invariant feature points. From the obtained results, the performance of the image registration is evaluated and it is concluded that the proposed approach is efficient. Copyright 2014 ACM.
暂无评论