Building edge or boundary extraction is always one of the most important issues for earth observation, city planning, and other applications. However, for accurately extracting building edge, there are commonly two difficult challenges. Firstly, unwanted strong edges from road and other things can be hardly avoided to be recognized. Secondly, it is more serious that many low or very low contrast weak edges will be not detected. In order to deal with these two issues to a certain extent, in this paper, based on sparse SVM with dual-scale features, we propose a Building Edge Extraction method in a Dual-scale Classification way with Decision Fusion embedded (DC-BEE). Specifically, with global linearity information as priori knowledge, training samples are selected automatically at first. Next, a sparse SVM classifier is trained using the dual-scale local edge features of the training samples. And then, the trained sparse SVM is employed to classify all extracted edges. Finally, the dual-scale decision fusion strategy is performed for final building edge extraction. Visual analysis and quantitative analysis of the experimental results from different style city regions illustrated that the proposed DC-BEE method can efficiently fulfill the building edge extraction task automatically.
Hyperspectral imaging has been widely applied in many fields due to the advantage of high spectral resolution. However, consisting of acquisition, transmission, reception and display, a hyperspectral imaging system may be disturbed in each part and thus leads to degradations that limit the precision of subsequent processing. It is therefore an important preprocessing step to remove the noise of acquired image data as much as possible. In this paper, we propose a novel regularization method for hyperspectral image denoising. Firstly, the low-rank and sparsity constraints are jointly used to establish the regularization model. For each spectral band, the low-rank constraint is for exploiting inter-column/- row correlations, while the sparsity constraint aims to exploit intra-column correlations. Secondly, reweighed ℓ1 norm strategy, which solves a sequence of weighted norm optimization problems and updates the weights with the solution ℓ1 of the last iteration, is introduced to approximate norm to achieve improved priori performance of the two ℓ0 constraints. Lastly, we apply the alternating direction method (ADM) under the augmented Lagrangian multiplier (ALM) framework to solve the model efficiently. Both low-rank and sparsity priors are reweighted at each iteration to promote low-rankness and sparseness of the solution. The denoising effect of our method is tested on real hyperspectral image data with different noise level. The experiments demonstrate the practicality of our proposed method.
Object detection is a fundamental problem faced in remote sensing images analysis. Most of object detection methods mainly focus on single-source image and utilize single spectral or spatial information. Therefore, they are easily affected by illumination angle, brightness and the structure similar to the object. To overcome these defects, a novel object detection framework is proposed using superpixel segmentation and multisource features in multispectral and panchromatic images. During multisource feature extraction stage, the local region spectral information and the spatial information are extracted from multispectral and panchromatic patches respectively. Then, we embed these spectral features into spatial features to construct the new multisource features. During the detection stage, superpixel segmentation method is applied to extract candidate patches based on the superpixel centers from multisource images, which makes detection more efficient. Then, multisource features are also extracted from these candidate patches, which are input to SVM for detection. Experiments are implemented using two groups of the panchromatic and multispectral images by WorldView 2. The results indicated that, compared with single-source detection result, the proposed method can effectively improve the detection performance both on precision and recall rate.
The region of interest (ROI) extraction is of crucial importance in the preprocessing of object detection, especially when the spatial resolution of the remote sensing image becomes extremely high and the field of view becomes relatively large. To conduct the detection approaches directly on the image usually yields unsatisfactory result, and is time consuming. Saliency models based on visual attention mechanism are the general solution to this problem. However, the conventional saliency models deal with the pixel intensity, color statistics or contrast, while neglect the characteristics and spatial distribution of the ROI, which would results in the false alarm in the extraction. In this paper, taken residential area as the region of interest, a ROI extraction method based on saliency, and enhanced by corner density is proposed. The saliency model is adopted to extract the potential area preliminarily. In spite of the efficiency of the model, it suffers from certain defect, that is, the preliminary extracted region contains plenty of false alarms due to the high contrast of bare land and water reflection. Therefore, corner density feature is constructed to refine the extraction, based on the idea of residential area showing higher edge and corner density compared to rural area. In the experimental part, the proposed method is compared with three saliency models. The experimental results reveal that the proposed method is effective in eliminating the false alarm caused by high intensity or contrast of the pixel.
Change detection techniques for remote sensing images are increasingly applied to many fields, such as disaster monitoring, vegetation coverage analysis and so on. With the increase of image spatial resolution, noises and details increased significantly, compared with the low-resolution image. How to improve the accuracy of change detection for high resolution image has been a critical topic. In this paper, a new method for high resolution image change detection based on pyramid mean shift smoothness and morphology is proposed. Firstly, the difference image is generated by fusing the difference feature and log difference feature based on stationary wavelet transform. Secondly, two-layer pyramid mean shift smoothness algorithm is applied to highlight the objects that may be changed and to eliminate interference regions, meanwhile, to retain the obviously different features. Thirdly, in order to enhance the contrast between the change objects and unchanged regions, the improved frequency-tuned saliency detection strategy is utilized to further enhance the change objects. Lastly, change objects are extracted by the fuzzy local C-means cluster algorithm and the final change map is generated by morphological operation. The method has been tested on four-temporal datasets, meanwhile, compared with other typical methods.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.