Ricardo Dutra da Silva

A novel feature descriptor based on the shearlet transform

Problems such as image classification, object detection and recognition rely on low-level feature descriptors to represent visual information. Several feature extraction methods have been proposed, including the Histograms of Oriented Gradients (HOG), which captures edge information by analyzing the distribution of intensity gradients and their directions. In addition to directions, the analysis of edge at different scales provides valuable information. Shearlet transforms provide a general framework for analyzing and representing data with anisotropic information at multiple scales. As a consequence, signal singularities, such as edges, can be precisely detected and located in images. Based on the idea of employing histograms to estimate the distribution of edge orientations and on the accurate multi-scale analysis provided by shearlet transforms, we propose a feature descriptor called Histograms of Shearlet Coefficients (HSC). Experimental results comparing HOG with HSC show that HSC provides significantly better results for the problems of texture classification and face identification.

Adaptive edge-preserving image denoising using wavelet transforms

Image denoising is a relevant issue found in diverse image processing and computer vision problems. It is a challenge to preserve important features, such as edges, corners and other sharp structures, during the denoising process. Wavelet transforms have been widely used for image denoising since they provide a suitable basis for separating noisy signal from the image signal. This paper describes a novel image denoising method based on wavelet transforms to preserve edges. The decomposition is performed by dividing the image into a set of blocks and transforming the data into the wavelet domain. An adaptive thresholding scheme based on edge strength is used to effectively reduce noise while preserving important features of the original image. Experimental results, compared to other approaches, demonstrate that the proposed method is suitable for different classes of images contaminated by Gaussian noise.

Satellite Image Segmentation Using Wavelet Transforms Based on Color and Texture Features

Image segmentation is a fundamental process in remote sensing applications, whose main purpose is to allow a meaningful discrimination among constituent regions of interest. This work presents a novel image segmentation method based on wavelet transforms for extracting a number of color and texture features from the images. Traditional feature extraction techniques based on individual pixels usually demand high computational cost. To reduce such computational cost, while achieving high-quality results, our approach is composed of two main stages. Initially, the image is decomposed into blocks of pixels and a wavelet transform is applied to each block to identify homogeneous regions of the image, assigning the entire block to a class. A refinement stage is applied to the remaining pixels which belong to blocks marked as heterogenous in the first stage. The developed method, tested on several remote sensing images and compared to a well known image segmentation method, presents high adaptability to image regions.

An alternative and smarter route planner for wheelchair users: Exploring open data

In this paper we describe a bottom-up approach to integrate GIS maps (endorsed by discrete features, such as points, lines, polygons), in order to develop a route planner for wheelchair users. We integrate public available data with a novel model for route planning, based on sidewalks, crosswalks and curb ramps, as opposed to traditional street-based approaches. We show that our sidewalk-based model is more suitable than available planning routes under mobility constraints, using a case study in Curitiba, Brazil.

Hyperspectral data classification improved by minimum spanning forests

Abstract. Remote sensing technology has applications in various knowledge domains, such as agriculture, meteorology, land use, environmental monitoring, military surveillance, and mineral exploration. The increasing advances in image acquisition techniques have allowed the generation of large volumes of data at high spectral resolution with several spectral bands representing images collected simultaneously. We propose and evaluate a supervised classification method composed of three stages. Initially, hyperspectral values and entropy information are employed by support vector machines to produce an initial classification. Then, the K-nearest neighbor technique searches for pixels with high probability of being correctly classified. Finally, minimum spanning forests are applied to these pixels to reclassify the image taking spatial restrictions into consideration. Experiments on several hyperspectral images are conducted to show the effectiveness of the proposed method.

3D Edge Detection Based on Boolean Functions and Local Operators

Edge detection is one of the most commonly used operations in image processing and computer vision areas. Edges correspond to the boundaries between regions in an image, which are useful for object segmentation and recognition tasks. This work presents a novel method for 3D edge detection based on Boolean functions and local operators, which is an extension of the 2D edge detector introduced by Vemis et al. [Signal Processing45(2), 161–172 (1995)] The proposed method is composed of two main steps. An adaptive binarization process is initially applied to blocks of the image and the resulting binary map is processed with a set of Boolean functions to identify edge points within the blocks. A global threshold, calculated to estimate image intensity variation, is then used to reduce false edges in the image blocks. The proposed method is compared to other 3D gradient filters: Canny, Monga–Deriche, Zucker–Hummel and Sobel operators. Experimental results demonstrate the effectiveness of the proposed technique when applied to several 3D synthetic and real data sets.

A Smarter Sidewalk-Based Route Planner for Wheelchair Users: An Approach with Open Data

In this chapter we describe an approach to integrate GIS maps (endorsed by discrete features, such as points, lines, polygons), in order to develop a route planner for wheelchair users. We integrate public available data and an approach with free software with a novel model for route planning, based on sidewalks, crosswalks and curb ramps, as opposed to traditional street-based approaches. We show that our sidewalk-based model is more suitable than available route planning services under mobility constraints, using a case study in Curitiba, Brazil.

A topology-based approach to computing neighborhood-of-interest points using the Morse complex

The work proposes a novel topological operator, called the Local Morse Context.It explores structural information in images through neighborhoods of interest points.The strategy avoids incorrect correspondences in high similarity images.The approach is designed and tested for the correspondence of stereo image pairs.Data sets with synthetic and real images are used in the evaluation of the method. A central problem in image processing and computer vision is the computation of corresponding interest points in a given set of images. Usually, interest points are considered as independent elements described by some local information. Due to the limitations of such an approach, many incorrect correspondences can be obtained. A specific contribution of this paper is the proposition of a topological operator, called Local Morse Context (LMC), computed over Morse complexes, introduced as a way of efficiently computing neighborhoods of interest points to explore the structural information in images. The LMC is used in the development of a matching algorithm, that helps reducing the number of incorrect matches, and obtaining a confidence measure of whether a correspondence is correct or incorrect. The approach is designed and tested for the correspondence of narrow-baseline synthetic and specially challenging underwater stereo pairs of images, for which traditional methods present difficulties for finding correct correspondences.

Constructing Point Clouds from Underwater Stereo Movies

Processing images of underwater environments of Antarctic lakes is challenging due to poor lighting conditions, low saturation and noise. This paper presents a novel pipeline for dense point cloud scene reconstruction from underwater stereo images and video obtained with low-cost consumer recording hardware. Features in stereo frames are selected and matched at high density. Putative matches are triangulated to produce point clouds in 3D space. Temporal feature tracking is performed to produce and merge point clouds. We demonstrate that this framework produces dense and accurate reconstructions for several tests.