PRARSHIA 2012 Abstracts

Full Papers
Full Papers
Paper Nr: 2
Title:

EVALUATION OF STEREO MATCHING COSTS ON CLOSE RANGE, AERIAL AND SATELLITE IMAGES
Authors:

Ke Zhu, Pablo d’Angelo and Matthias Butenuth
Abstract: In the last years, most dense stereo matching methods use evaluation on the Middlebury stereo vision benchmark datasets. Most recent stereo algorithms were designed to perform well on these close range stereo datasets with relatively small baselines and good radiometric behaviour. In this paper, different matching costs on the Semi-Global Matching algorithm are evaluated and compared using the common Middlebury datasets, aerial and satellite datasets with ground truth. The experimental results show that the performance of dense stereo methods for datasets with larger baselines and stronger radiometric changes relies on even more robust matching costs. In addition, a novel matching cost based on mutual information and Census is introduced showing the most robust performance on close range, aerial and satellite data.
Download
Paper Nr: 6
Title:

NON-INVASIVE MELANOMA DIAGNOSIS USING MULTISPECTRAL IMAGING
Authors:

Ianisse Quinzán Suárez, Pedro Latorre Carmona, Pedro García Sevilla, Enrique Boldo, Filiberto Pla, Vicente García Jiménez, Rafael Lozoya and Guillermo Pérez de Lucía
Abstract: The early analysis of pigmented skin lesions is important for clinicians in order to recognize malignant melanoma. However, it is difficult to differentiate it from benign skin lesions due to their similarity based on their appearance. Since melanoma has a tendency to grow inside the skin and the depth of penetration of light into the skin is wavelength dependent, a multispectral imaging acquisition and processing approach to classify pigmented lesions as melanoma seems appropriate. This paper presents a method to diagnose melanoma lesions over a group of 26 samples acquired with a multispectral system, where 6 of them are melanomas, and the other 20 are other types of pigmented lesions. A Leave-One-Out strategy is used to create the training/test set. The classification imbalance problem inherent to this dataset is alleviated using a SMOTE technique. The random component of the SMOTE methodology is dealt with running it 25 times and a Qualified Majority Voting (QMV) scheme is used to do the final classification, using SVM. Results show this strategy allows to obtain competitive classification quality results.
Download
Paper Nr: 7
Title:

FUSION OF OPTICAL AND THERMAL IMAGERY AND LIDAR DATA FOR APPLICATION TO 3-D URBAN ENVIRONMENT - Pattern Recognition Applications in Remotely Sensed Hyperspectral Image Analysis
Authors:

Anna Brook, Marijke Vandewal, Rudolf Richter and Eyal Ben-Dor
Abstract: Investigation of urban environment includes a wide range of applications that require 3-D information. New approaches are needed for near-real-time analysis of the urban environment with natural 3-D visualization of extensive coverage. The remote sensing technology is a promising and powerful tool to assess quantitative information of urban materials and structures. This technique provides ability for easy, rapid and accurate in situ assessment of corrosion, deformations and ageing processes in the spatial (2-D) and the spectral domain within near-real-time and with high temporal resolution. LiDAR technology offers precise information about the geometrical properties of the surfaces and can reflect the different shapes and formations in the complex urban environment. Generating a monitoring system that is based on integrative fusion of hyperspectral, thermal and LiDAR data may enlarge the application envelope of each individual technology and contribute valuable information on the built urban environment. A fusion process defined by a data-registration algorithm and including spectral/thermal/spatial and 3-D information has been developed. The proposed practical 3-D urban environment application may provide urban planners, civil engineers and decision-makers with tools to consider temporal, quantitative and thermal spectral information in the 3-D urban space.
Download
Paper Nr: 9
Title:

A NEW MULTIPLE CLASSIFIER SYSTEM FOR SEMI-SUPERVISED ANALYSIS OF HYPERSPECTRAL IMAGES
Authors:

Jun Li, Prashanth Reddy Marpu, Antonio Plaza, Jose Manuel Bioucas Dias and Jon Atli Benediktsson
Abstract: In this work, we propose a new semi-supervised algorithm for remotely sensed hyperspectral image classification which belongs to the family of multiple classifier systems. The proposed approach combines the output of two well-established discriminative classifiers: sparse multinomial logistic regression (SMLR) and quadratic discriminant analysis (QDA). Our approach follows a two-step strategy. First, both SMLR and QDA are trained from the same set of labeled training samples and make predictions for the unlabeled samples in the image. Second, the set of unlabeled training samples is expanded by combining the estimates obtained by both classifiers in the previous step. The effectiveness of the proposed method is evaluated via experiments with a widely used hyperspectral image, collected by the Airborne Visible Infra-Red Imaging Spectrometer (AVIRIS) over the Indian Pines region in Indiana. Our results indicate that the proposed multiple classifier method provides state-of-the-art performance when compared to other methods.
Download
Paper Nr: 10
Title:

AUTOMATIC SELECTION OF THE TRAINING SET FOR SEMI-SUPERVISED LAND CLASSIFICATION AND SEGMENTATION OF SATELLITE IMAGES
Authors:

Olga Rajadell and Pedro García Sevilla
Abstract: Different scenarios can be found in land classification and segmentation of satellite images. First, when prior knowledge is available, the training data is generally selected by randomly picking samples within classes. When no prior knowledge is available the system can pick samples at random among all unlabeled data, which is highly unreliable, and ask the expert to label them or it can rely on the expert collaboration to improve progressively the training data applying an active learning function. We suggest a scheme to tackle the lack of prior knowledge without actively involving the expert, whose collaboration may be expensive. The proposed scheme uses a clustering technique to analyze the feature space and find the most representative samples for being labeled. In this case the expert is just involved in labeling once a reliable training data set for being representative of the features space. Once the training set is labeled by the expert, different classifiers may be built to process the rest of samples. Three different approaches are presented in this paper: the result of the clustering process, a distance based classifier, and support vector machines (SVM).
Download
Paper Nr: 11
Title:

NEURAL BASED ROTATION AND SCALE INDEPENDENT DETECTION OF TARGETS IN A HYPERSPECTRAL WATERWAY MONITORING SYSTEM
Authors:

Blanca Priego, Richard J. Duro, Francisco Bellas and Daniel Souto
Abstract: This paper is devoted to the presentation of the orientation and scale invariant detection subsystem within the current development of Hywacoss (Hyperspectral waterway control and security system). A neural network ensemble based identification and rotation detection module is considered in order to be able to detect and classify objects in waterways from hyperspectral image cubes in a fast and efficient manner. The neural approach followed is inspired by the orientation detection structures in the visual processing cortex. The system is tested over two different hyperspectral image cubes extracted from simulated waterways to verify its adequate operation.
Download
Paper Nr: 12
Title:

DICTIONARY BASED HYPERSPECTRAL IMAGE RETRIEVAL
Authors:

Miguel A. Veganzones, Mihai Datcu and Manuel Graña
Abstract: The normalized information distance (NID) is an universal metric distance based on Kolmogorov complexity. However, NID is not computable in a Turing sense. The normalized compression distance (NCD) is a computable distance that approximates NID by using normal compressors. NCD is a parameter-free distance that compares two signals by their lengths after separate compression relative to the length of the signal resulting from their concatenation after compression. The use of NCD for image retrieval over large image databases is difficult due to the computational cost of compressing the query image concatenated with every image in the database. The use of dictionaries extracted by dictionary-based compressors, such as the LZW compression algorithm, has been proposed to overcome this problem. Here we propose a Content-Based Image Retrieval system based on such dictionaries for the mining of hyperspectral databases. We compare results using the Normalized Dictionary Distance (NDD) and the Fast Dictionary Distance (FDD) against the NCD over different datasets of hyperspectral images. Results validate the applicability of dictionaries for hyperspectral image retrieval.
Download
Paper Nr: 14
Title:

ABOUT GRADIENT OPERATORS ON HYPERSPECTRAL IMAGES
Authors:

Ramón Moreno and Manuel Graña
Abstract: Gradient operators allow image segmentation based on edge information. Gradient operators based on chromatic information may avoid apparent edges detection due to illumination effects. This paper proposes the extension of chromatic gradients defined for RGB color images to images with n-dimensional pixels. A spherical coordinate representation of the pixel's content provides the required chromatic information. The paper provides results showing that gradient operators defined on the spherical coordinate representation effectively avoid illumination induced false edge detection.
Download
Paper Nr: 16
Title:

HYPERSPECTRAL UNMIXING WITH SIMULTANEOUS DIMENSIONALITY ESTIMATION
Authors:

Jose M. P. Nascimento and José M. Bioucas-Dias
Abstract: This paper is an elaboration of the simplex identification via split augmented Lagrangian (SISAL) algorithm (Bioucas-Dias, 2009) to blindly unmix hyperspectral data. SISAL is a linear hyperspectral unmixing method of the minimum volume class. This method solve a non-convex problem by a sequence of augmented Lagrangian optimizations, where the positivity constraints, forcing the spectral vectors to belong to the convex hull of the endmember signatures, are replaced by soft constraints. With respect to SISAL, we introduce a dimensionality estimation method based on the minimum description length (MDL) principle. The effectiveness of the proposed algorithm is illustrated with simulated and real data.
Download