The availability of huge remote sensing image dataset imposes recourse to powerful techniques of ... more The availability of huge remote sensing image dataset imposes recourse to powerful techniques of content-based image retrieval for archiving and mining. This paper propose descriptors based on the SIFT (Scale invariant features) combined with SVM linear classification. To build a powerful image classifier using very little training data, image augmentation is usually required to boost the performance of the classification. For this reason, an augmentation data is used to increase the training data for the SVM (Support vector Machine) classifier. The creation of the training data is done using several techniques of augmentation: anisotropic filter. We report a first evaluation of the CBIR (Content based image retrieval) and the second evaluation of the system aims to compare the deep learning with the boosted SVM classification.
Abstract. This paper describes a central processing unit (CPU)-based technique for terrain geomet... more Abstract. This paper describes a central processing unit (CPU)-based technique for terrain geometry rendering that could relieve graphics processing unit (GPU) from processing the appropriate level of detail (LOD) of the geometric surface. The proposed approach alleviates the computational load on the CPU and approaches GPU-based efficiency. As the datasets of realistic terrains are usually huge for real-time rendering, we suggest using a training stage to handle large tiled QuadTree terrain representation. The training stage is based on multiresolution wavelet decomposition and is used to limit the region of error control inside the tile. Maximum approximation errors are then calculated for each tile at different resolutions. Maximum world-space errors of the tile at different resolutions permit selection of the appropriate resolution of downsampling that will represent the tile at the run time. Tests and experiments demonstrate that B-spline 0 and B-spline 1 wavelets, well known for their properties of localization and their compact support, are suitable for fast and accurate localization of the maximum approximation error. The experimental results demonstrate that the proposed approach drastically reduces computation time in the CPU. Such a technique should also be used on low/medium end PCs, and embedded systems that are not equipped with the latest models of graphic hardware.
In this paper we propose a new rotation and scale invariant representation for high resolution sa... more In this paper we propose a new rotation and scale invariant representation for high resolution satellite image retrieval based on Steerable Pyramid Decomposition. By calculating the statistical measures of decomposed image subbands, the texture feature vectors are extracted. To obtain rotation and scale invariance, the feature vectors are circularly shifted until obtaining the minimum possible distance. Experiments were conducted using 8 image classes from land-use/land-cover (LULC) UCMerced dataset. Results obtained are compared with color Gabor opponent texture features. Tests and evaluation measures demonstrate that the proposed technique gives a good performance in terms of high precision.
In this paper, we evaluate the performance of SURF descriptor for high resolution satellite image... more In this paper, we evaluate the performance of SURF descriptor for high resolution satellite imagery (HRSI) retrieval through a BoVW model on a land-use/land-cover (LULC) dataset. Local feature approaches such as SIFT and SURF descriptors can deal with a large variation of scale, rotation and illumination of the images, providing, therefore, a better discriminative power and retrieval efficiency than global features, especially for HRSI which contain a great range of objects and spatial patterns. Moreover, we combine SURF and color features to improve the retrieval accuracy, and we propose to learn a category-specific dictionary for each image category which results in a more discriminative image representation and boosts the image retrieval performance.
In this paper, we propose a novel method of optimizing texture primitives description and segment... more In this paper, we propose a novel method of optimizing texture primitives description and segmentation using variography. Indeed, successful textural analysis relies on the careful selection of the adapted window size. The variogram is essentially a "variance of differences" in the values as a function of the separation distance. This variance therefore changes as the separation distance increases where repetitive
Abstract. A local-region approach based on bag-of-visual-words model for high-resolution satellit... more Abstract. A local-region approach based on bag-of-visual-words model for high-resolution satellite image (HRSI) retrieval and classification is proposed. The proposed method effectively describes HRSI and, hence, considerably reduces the semantic gap. The local representation is achieved by an almost complete description of key points, through the proposed color-texture-structure-spectral-speeded-up robust features (CTSS-SURF) descriptor. The CTSS-SURF can effectively overcome the challenges of HRSI, such as scale, illumination, shift, and rotation variation. The regional representation is achieved by dividing images into several parts and then designing regional feature vectors. For both representations, an improved procedure for dictionary creation is proposed to increase the dictionary discriminative ability and reduce the computational cost. An extensive experimental evaluation on the UC Merced Land Use Dataset has been performed and compared with other feature extraction methods. For retrieval, the proposed method achieves 81.24 in mean average precision value and an accuracy of 98.23% for scene classification. It outperforms many state-of-the-art methods, including convolutional neural networks. The impressive results demonstrate the superiority of the proposed approach for HRSI.
With the steadily expanding demand for remote sensing images, many satellites have been launched,... more With the steadily expanding demand for remote sensing images, many satellites have been launched, and thousands of high resolution satellite images (HRSI) are acquired every day. Therefore, retrieving useful images quickly and accurately from a huge image database has become a challenge. In this paper, we propose an adaptive content-based image retrieval (CBIR) system for the retrieval of HRSI on the basis of Steerable Pyramids using RGB and CIElab color systems. The texture feature vectors are extracted by calculating the statistical measures of decomposed image sub-bands. To improve the performances of our CBIR scheme, the system rotation and scale invariance is enhanced by introducing a circular shifting of the feature vector elements according to each scale. Extensive experiments were conducted firstly using 8 image classes from land-use/land-cover (LULC) UCMerced dataset. Obtained results are compared with color Gabor opponent texture features. The system was then extended to work on the whole dataset consisting of 21 image classes, and compared with results obtained from SIFT descriptor. The tests and evaluation measures demonstrate that the proposed system gives a good performance in terms of high precision.
Abstract. This paper highlights color component features that improve high-resolution satellite (... more Abstract. This paper highlights color component features that improve high-resolution satellite (HRS) images retrieval. Color component correlation across image lines and columns is used to define a revised color space. It is designed to simultaneously take both color and neighborhood information. From this space, color descriptors, namely rotation invariant uniform local binary pattern, histogram of gradient, and a modified version of local variance are derived through dual-tree complex wavelet transform (DT-CWT). A new color descriptor called smoothed local variance (SLV) using an edge-preserving smoothing filter is introduced. It is intended to offer an efficient way to represent texture/structure information using an invariant to rotation descriptor. This descriptor takes advantage of DT-CWT representation to enhance the retrieval performance of HRS images. We report an evaluation of the SLV descriptor associated with the new color space using different similarity distances in our content-based image retrieval scheme. We also perform comparison with some standard features. Experimental results show that SLV descriptor allied to DT-CWT representation outperforms the other approaches.
This paper reports investigations aimed at developing a variographic tool which visually describe... more This paper reports investigations aimed at developing a variographic tool which visually describes or discriminates perceivable qualities of textures. In particular, it will be shown that the variogram measure commonly used in geostatistic can be used to identify different structures. In texture analysis the first and most important task is to extract texture features which most completely embody information about
In this paper, we evaluate and compare the performance of three machine learning classifiers: Sup... more In this paper, we evaluate and compare the performance of three machine learning classifiers: Support Vector Machines (SVM), Decision Trees (DT) and K-Nearest Neighbor (K-NN) for high resolution satellite image scene classification.This study aims at providing insights into the selection of the appropriate classifier and highlighting the importance of the appropriate setting of the classifier parameters. We illustrate these issues through applying scene classification to UC-Merced high resolution satellite image dataset. Image features are obtained through the SURF descriptor and BOVW model.
The availability of huge remote sensing image dataset imposes recourse to powerful techniques of ... more The availability of huge remote sensing image dataset imposes recourse to powerful techniques of content-based image retrieval for archiving and mining. This paper propose descriptors based on the SIFT (Scale invariant features) combined with SVM linear classification. To build a powerful image classifier using very little training data, image augmentation is usually required to boost the performance of the classification. For this reason, an augmentation data is used to increase the training data for the SVM (Support vector Machine) classifier. The creation of the training data is done using several techniques of augmentation: anisotropic filter. We report a first evaluation of the CBIR (Content based image retrieval) and the second evaluation of the system aims to compare the deep learning with the boosted SVM classification.
Abstract. This paper describes a central processing unit (CPU)-based technique for terrain geomet... more Abstract. This paper describes a central processing unit (CPU)-based technique for terrain geometry rendering that could relieve graphics processing unit (GPU) from processing the appropriate level of detail (LOD) of the geometric surface. The proposed approach alleviates the computational load on the CPU and approaches GPU-based efficiency. As the datasets of realistic terrains are usually huge for real-time rendering, we suggest using a training stage to handle large tiled QuadTree terrain representation. The training stage is based on multiresolution wavelet decomposition and is used to limit the region of error control inside the tile. Maximum approximation errors are then calculated for each tile at different resolutions. Maximum world-space errors of the tile at different resolutions permit selection of the appropriate resolution of downsampling that will represent the tile at the run time. Tests and experiments demonstrate that B-spline 0 and B-spline 1 wavelets, well known for their properties of localization and their compact support, are suitable for fast and accurate localization of the maximum approximation error. The experimental results demonstrate that the proposed approach drastically reduces computation time in the CPU. Such a technique should also be used on low/medium end PCs, and embedded systems that are not equipped with the latest models of graphic hardware.
In this paper we propose a new rotation and scale invariant representation for high resolution sa... more In this paper we propose a new rotation and scale invariant representation for high resolution satellite image retrieval based on Steerable Pyramid Decomposition. By calculating the statistical measures of decomposed image subbands, the texture feature vectors are extracted. To obtain rotation and scale invariance, the feature vectors are circularly shifted until obtaining the minimum possible distance. Experiments were conducted using 8 image classes from land-use/land-cover (LULC) UCMerced dataset. Results obtained are compared with color Gabor opponent texture features. Tests and evaluation measures demonstrate that the proposed technique gives a good performance in terms of high precision.
In this paper, we evaluate the performance of SURF descriptor for high resolution satellite image... more In this paper, we evaluate the performance of SURF descriptor for high resolution satellite imagery (HRSI) retrieval through a BoVW model on a land-use/land-cover (LULC) dataset. Local feature approaches such as SIFT and SURF descriptors can deal with a large variation of scale, rotation and illumination of the images, providing, therefore, a better discriminative power and retrieval efficiency than global features, especially for HRSI which contain a great range of objects and spatial patterns. Moreover, we combine SURF and color features to improve the retrieval accuracy, and we propose to learn a category-specific dictionary for each image category which results in a more discriminative image representation and boosts the image retrieval performance.
In this paper, we propose a novel method of optimizing texture primitives description and segment... more In this paper, we propose a novel method of optimizing texture primitives description and segmentation using variography. Indeed, successful textural analysis relies on the careful selection of the adapted window size. The variogram is essentially a "variance of differences" in the values as a function of the separation distance. This variance therefore changes as the separation distance increases where repetitive
Abstract. A local-region approach based on bag-of-visual-words model for high-resolution satellit... more Abstract. A local-region approach based on bag-of-visual-words model for high-resolution satellite image (HRSI) retrieval and classification is proposed. The proposed method effectively describes HRSI and, hence, considerably reduces the semantic gap. The local representation is achieved by an almost complete description of key points, through the proposed color-texture-structure-spectral-speeded-up robust features (CTSS-SURF) descriptor. The CTSS-SURF can effectively overcome the challenges of HRSI, such as scale, illumination, shift, and rotation variation. The regional representation is achieved by dividing images into several parts and then designing regional feature vectors. For both representations, an improved procedure for dictionary creation is proposed to increase the dictionary discriminative ability and reduce the computational cost. An extensive experimental evaluation on the UC Merced Land Use Dataset has been performed and compared with other feature extraction methods. For retrieval, the proposed method achieves 81.24 in mean average precision value and an accuracy of 98.23% for scene classification. It outperforms many state-of-the-art methods, including convolutional neural networks. The impressive results demonstrate the superiority of the proposed approach for HRSI.
With the steadily expanding demand for remote sensing images, many satellites have been launched,... more With the steadily expanding demand for remote sensing images, many satellites have been launched, and thousands of high resolution satellite images (HRSI) are acquired every day. Therefore, retrieving useful images quickly and accurately from a huge image database has become a challenge. In this paper, we propose an adaptive content-based image retrieval (CBIR) system for the retrieval of HRSI on the basis of Steerable Pyramids using RGB and CIElab color systems. The texture feature vectors are extracted by calculating the statistical measures of decomposed image sub-bands. To improve the performances of our CBIR scheme, the system rotation and scale invariance is enhanced by introducing a circular shifting of the feature vector elements according to each scale. Extensive experiments were conducted firstly using 8 image classes from land-use/land-cover (LULC) UCMerced dataset. Obtained results are compared with color Gabor opponent texture features. The system was then extended to work on the whole dataset consisting of 21 image classes, and compared with results obtained from SIFT descriptor. The tests and evaluation measures demonstrate that the proposed system gives a good performance in terms of high precision.
Abstract. This paper highlights color component features that improve high-resolution satellite (... more Abstract. This paper highlights color component features that improve high-resolution satellite (HRS) images retrieval. Color component correlation across image lines and columns is used to define a revised color space. It is designed to simultaneously take both color and neighborhood information. From this space, color descriptors, namely rotation invariant uniform local binary pattern, histogram of gradient, and a modified version of local variance are derived through dual-tree complex wavelet transform (DT-CWT). A new color descriptor called smoothed local variance (SLV) using an edge-preserving smoothing filter is introduced. It is intended to offer an efficient way to represent texture/structure information using an invariant to rotation descriptor. This descriptor takes advantage of DT-CWT representation to enhance the retrieval performance of HRS images. We report an evaluation of the SLV descriptor associated with the new color space using different similarity distances in our content-based image retrieval scheme. We also perform comparison with some standard features. Experimental results show that SLV descriptor allied to DT-CWT representation outperforms the other approaches.
This paper reports investigations aimed at developing a variographic tool which visually describe... more This paper reports investigations aimed at developing a variographic tool which visually describes or discriminates perceivable qualities of textures. In particular, it will be shown that the variogram measure commonly used in geostatistic can be used to identify different structures. In texture analysis the first and most important task is to extract texture features which most completely embody information about
In this paper, we evaluate and compare the performance of three machine learning classifiers: Sup... more In this paper, we evaluate and compare the performance of three machine learning classifiers: Support Vector Machines (SVM), Decision Trees (DT) and K-Nearest Neighbor (K-NN) for high resolution satellite image scene classification.This study aims at providing insights into the selection of the appropriate classifier and highlighting the importance of the appropriate setting of the classifier parameters. We illustrate these issues through applying scene classification to UC-Merced high resolution satellite image dataset. Image features are obtained through the SURF descriptor and BOVW model.
Uploads