By definition, HSC (HyperSpectral Camera) images are much richer in spectral data than, say, a COTS (Commercial-Off-The-Shelf) color camera. But data are not information. If we do the task right, useful information ca...
详细信息
By definition, HSC (HyperSpectral Camera) images are much richer in spectral data than, say, a COTS (Commercial-Off-The-Shelf) color camera. But data are not information. If we do the task right, useful information can be derived from the data in HSC images. Nature faced essentially the identical problem. The incident light is so complex spectrally that measuring it with high resolution would provide far more data than animals can handle in real time. Nature's solution was to do irreversible POCS (Projections Onto Convex Sets) to achieve huge reductions in data with minimal reduction in information. Thus we can arrange for our manmade systems to do what nature did - project the HSC image onto two or more broad, overlapping curves. The task we have undertaken in the last few years is to develop this idea that we call Artificial Color. What we report here is the use of the measured HSC image data projected onto two or three convex, overlapping, broad curves in analogy with the sensitivity curves of human cone cells. Testing two quite different HSC images in that manner produced the desired result: good discrimination or segmentation that can be done very simply and hence are likely to be doable in real time with specialized computers. Using POCS on the HSC data to reduce the processing complexity produced excellent discrimination in those two cases. For technical reasons discussed here, the figures of merit for the kind of patternrecognition we use is incommensurate with the figures of merit of conventional patternrecognition. We used some force fitting to make a comparison nevertheless, because it shows what is also obvious qualitatively. In our tasks our method works better.
In this paper we introduce a probabilistic approach of building extraction in remotely sensed images. To cope with data heterogeneity we construct a flexible hierarchical framework which can create various building ap...
详细信息
ISBN:
(纸本)9781424475421
In this paper we introduce a probabilistic approach of building extraction in remotely sensed images. To cope with data heterogeneity we construct a flexible hierarchical framework which can create various building appearance models from different elementary feature based modules. A global optimization process attempts to find the optimal configuration of buildings, considering simultaneously the observed data, prior knowledge, and interactions between the neighboring building parts. The proposed method is evaluated on various aerial image sets containing more than 500 buildings, and the results are matched against two state-of-the-art techniques.
To extract information at the individual tree level, which is very useful in biology, ecology and forestry, would be prohibitively time-consuming and be necessary for artificial intelligence by considering many factor...
详细信息
To extract information at the individual tree level, which is very useful in biology, ecology and forestry, would be prohibitively time-consuming and be necessary for artificial intelligence by considering many factors. The presented approach develops a tree top seeded based region growth tree detection and crown delineation algorithm for analyzing QuickBird satellite images in Populus × xiaohei plantation even stand at Xue Jia Zhuang wood farm in Shanxi Province of China. After multi resolution segmentation, we get image object segments for tree top seeds detection with NDVI and ratio NIR feature. Around theses seeds, we let them region growing in a cycle way. Some false seeds must be wiped off with given feature threshold. After quad tree segmentation for crown shape optimization, the same category region must be merged. We use 9 plots with different plantation density to validate the above method. Average tree numbers identification error is 18.9%, R 2 = 0.4693. From comparing tree numbers of field work and software identification by tree matching, the confusion matrix, overall accuracy, commission error, omission error is computed.
High resolution still image compression and storage plays a very important role in the application field of aero military reconnaissance, topography mapping and satellite remotesensing exploration. An embedded high r...
详细信息
High resolution still image compression and storage plays a very important role in the application field of aero military reconnaissance, topography mapping and satellite remotesensing exploration. An embedded high resolution still image compression and storage system based on POWER PC, FPGA and ADV212 is presented in the paper in order to implement 4008×5366×12bit lossy and lossless compression of high resolution aero image. image compression storage system structure and the software and hardware design method of the system are shown here. The design object is to save one image each two seconds and to send one image each eight seconds on the condition of 1.44Mbps transmission rate. The experimental results show that the system could implement expected real-time compression and storage of high resolution aero image.
The traditional Mallat's algorithm uses an orthogonal wavelet basis, but the transform is not shift-invariant, and the Mallat's algorithm has a major disadvantage for it contains interpolation and decimation, ...
详细信息
The traditional Mallat's algorithm uses an orthogonal wavelet basis, but the transform is not shift-invariant, and the Mallat's algorithm has a major disadvantage for it contains interpolation and decimation, which can be a problem in signal analysis or patternrecognition. Moreover, with the increasing of decomposition level, the loss of spectrum becomes more serious. So in this paper, a new method that combines local regional features fusion rule with traditional HSI - À Trous algorithm to fuse the ETM+ remotesensingimages is proposed. By comparing its result with HSI-DWT algorithm as well as traditional À Trous algorithm, the results indicate the new method can not only preserve spatial details information better, but also greatly reduce the spectrum loss of the multi-spectral images.
Support Vector Machines is a supervised classifier which used kernel functions to mitigate nonlinear problem. Various kernel functions like Gaussian and polynomial kernels previously used for hyperspectral image class...
详细信息
ISBN:
(纸本)9781424455690
Support Vector Machines is a supervised classifier which used kernel functions to mitigate nonlinear problem. Various kernel functions like Gaussian and polynomial kernels previously used for hyperspectral image classification. In this paper, new kernel function is used for hyperspectral image classification. This kernel is based on wavelet which named wavelet-kernel. The comparative result of Wavelet kernel with two common kernels are given which shows wavelet kernel is a good choice for SVM classifier in remotesensing.
Car-plate locating is the first step of the recognition of car-plate,the accuracy of locating plays an important role in the next *** paper introduce some methods of car-plate ***,the input image is roughly segmented ...
详细信息
Car-plate locating is the first step of the recognition of car-plate,the accuracy of locating plays an important role in the next *** paper introduce some methods of car-plate ***,the input image is roughly segmented absed on the color information,and some regions with the color of car-plate are *** the false regions in the result image is *** last,projection method is used to get the accurate positon of the *** experimental result shows that the proposed method is very effective.
It is becoming study hotspot of academia that using visible 3D expression of face curved surface to recognize face. These issues involved in founding 3D-face recognition system were analyzed such as basic process, sys...
详细信息
It is becoming study hotspot of academia that using visible 3D expression of face curved surface to recognize face. These issues involved in founding 3D-face recognition system were analyzed such as basic process, system function, technology line, system architecture and database programming. Then some key technologies were explored which were development of management platform, analysis of video frequency flow, automatic demarcation of character points on the face, calculation of shooting angle and so on.
Desert roads have casual dust events in which the dust may completely cover several kilometers of the road for several hours and this usually causes breathing problems and may deadly accidents for travelers. The syste...
详细信息
Desert roads have casual dust events in which the dust may completely cover several kilometers of the road for several hours and this usually causes breathing problems and may deadly accidents for travelers. The system is based on a real-time analysis to classify the activity of the dust of the acquired images. The analysis is performed by processing the images within the wireless nodes and extracting information about the dust events. Recently wireless image sensor networks have become of a great importance due to their wide application range such as video surveillance, remotesensing, tracking, face recognition and so on. Desert roads have casual dust storm events in which the dust may completely cover several kilometers of the road for several hours and this usually causes breathing problems and accidents for travelers.
The 2-D Gabor transform has been recognized as being useful in diverse areas such as image compression, texture analysis, image segmentation, and imagerecognition; however, its real time applications have been limite...
详细信息
The 2-D Gabor transform has been recognized as being useful in diverse areas such as image compression, texture analysis, image segmentation, and imagerecognition; however, its real time applications have been limited due to the high computational complexity of the traditional 2-D complex-valued discrete Gabor transform (CDGT) algorithms. In this paper, 2-D discrete Hartley transform (DHT)-based real-valued discrete Gabor transform (RDGT) is presented, which can utilize the fast DHT algorithm for fast computation and has a simple relationship with the CDGT such that the 2-D CDGT coefficients can be directly computed from the 2-D DHT-based RDGT coefficients. Due to this simple relationship, the 2-D DHT-based RDGT also offers a faster and more efficient method to compute the 2-D CDGT. In addition, an efficient algorithm for the fast computation of the biorthogonal analysis window given a synthesis window is also presented. The results indicate that the proposed algorithms for the RDGT are attractive for real time imageprocessing.
暂无评论