Leyza Baldo Dorini

White blood cell segmentation using morphological operators and scale-space analysis

This paper proposes a model-based methodology for recognizing and tracking objects in digital image sequences. Objects are represented by attributed relational graphs (or ARGs), which carry both local and relational information about them. The recognition is performed by inexact graph matching, which consists in finding an approximate homomorphism between ARGs derived from an input video and a model image. Searching for a suitable homomorphism is achieved through a tree-search optimization algorithm and the minimization of a pre-defined cost function. Motion smoothness between successive frames is exploited to achieve the recognition over the whole sequence, with improved spatio-temporal coherence.Cell segmentation is a challenging problem due to both the complex nature of the cells and the uncertainty present in video microscopy. Manual methods for this purpose are onerous, imprecise and highly subjective, thus requiring automated methods that perform this task in an objective and efficient way. In this paper, we propose a novel method to segment nucleus and cytoplasm of white blood cells (WBC). WBC composition of the blood provides important information to doctors and plays an important role in the diagnosis of different diseases. We use simple morphological operators and explore the scale-space properties of a toggle operator to improve the segmentation accuracy. The proposed scheme has been successfully applied to a large number of images, showing promising results for varying cell appearance and image quality, encouraging future works.

Semiautomatic White Blood Cell Segmentation Based on Multiscale Analysis

This paper approaches novel methods to segment the nucleus and cytoplasm of white blood cells (WBC). This information is the basis to perform higher level tasks such as automatic differential counting, which plays an important role in the diagnosis of different diseases. We explore the image simplification and contour regularization resulting from the application of the selfdual multiscale morphological toggle (SMMT), an operator with scale-space properties. To segment the nucleus, the image preprocessing with SMMT has shown to be essential to ensure the accuracy of two well-known image segmentations techniques, namely, watershed transform and Level-Set methods. To identify the cytoplasm region, we propose two different schemes, based on granulometric analysis and on morphological transformations. The proposed methods have been successfully applied to a large number of images, showing promising segmentation and classification results for varying cell appearance and image quality, encouraging future works.

A scaled morphological toggle operator for image transformations

Scale dependent signal representations have proved to be useful in several image processing applications. In this paper, we define a toggle operator for binarization/segmentation purposes based on scaled versions of an image transformed by morphological operations. The toggle decision rule, determining the new value of a pixel, considers local spatial information, in contrast to other multiscale approaches that takes into account mainly global information (e.g., the scale signal under study). We show that the proposed operator can identify significant image extrema information in such a way that when it is used in a binarization process yields very good segmentation and filtering results. Our algorithm is validated against known threshold-based segmentation methods using images of different classes and subjected to different lighting conditions

Multiscale Morphological Image Simplification

Image simplification reduces the information content of an image, being frequently used as a preprocessing stage in several algorithms to suppress undesired details such as noise. Morphological filters, commonly used for this purpose, have as main drawbacks the asymmetric treatment of peaks and valleys and the difficulty to choose an appropriate structuring element size. Here, we propose a self-dual multiscale image simplification operator with sound edge preservation properties. This enables us to represent the inherent multiscale nature of real-world images by embedding the original signal into a family of derived signals, which represent simplified versions of the image obtained by successively removing its structures across scales. Thus, it is possible to analyze the different representation levels to extract the interest features, and the definition of a structure element size does not constitute a problem anymore. Based on these notions, we present some experiments on image segmentation, a basic step of various pattern recognition approaches.

A SELF-DUAL FILTERING TOGGLE OPERATOR FOR SPECKLE NOISE FILTERING

In this work, we formalize an alternative way to build self-dual morphological filters that extends some results obtained for morphological centers to a different class of toggle operators. Thus, a wider range of primitives can be considered without causing oscillations, a common problem in toggle mappings. We also show that the combination of the morphological filters generated by using the proposed approach with the well-known anisotropic diffusion technique yields sound results where homogeneous regions are smoothed without degrading edge information. We explore the filtering of speckle noise, an interference pattern that causes a granular aspect in the image, thus limiting its interpretation and making it difficult further image processing tasks. Experimental tests on both synthetic and real-world images show promising results when compared to some well-known methods related to this type of filtering.

A Multiscale Morphological Binarization Algorithm

Image binarization is widely used to generate more appropriate images to be used in several image analysis and understanding systems, as well as to facilitate data management and decrease storage space requirements. The main difficulties arise from the fact that images are frequently degraded by noise and non-uniform illumination, for example. This paper presents an efficient morphological-based image binarization technique with scale-space properties that is able to cope with these problems. We evaluate the proposed approach for different classes of images, including text images such as historical and machine-printed documents, obtaining promising results.

Sentiment Analysis in Outdoor Images Using Deep Learning

In this work, we explore how Convolutional Neural Networks can be applied to the task of sentiment analysis in visual media. We compare four different architectures and propose a new approach where attributes that represent the main categories used for scenes description are combined with the output of the convolutional layers before the classification process. In the first dataset, composed of image tweets, we obtained accuracy improvements over previous works. The second dataset, constructed in this paper, contains only images from outdoor areas and labeled in three sentiment classes: positive, neutral and negative. Sentiment analysis of outdoor images helps to enable new services, e.g., to better uncover the semantics of areas compared to indoor images. In general, the use of the attributes improves the accuracy of the results.

Análise Comparativa de Métodos de Segmentação e Técnicas de Aprendizado de Máquina Com Aplicação no Reconhecimento Automático de Placas de Identificação de Veículos

Este trabalho trata do problema de Reconhecimento Automatico de Placas de Veiculos (do ingles, Automatic Number Plate Recognition - ANPR), o qual envolve o uso de diferentes tecnicas, incluindo Reconhecimento Optico de Caracteres (do ingles, Optical Character Recognition - OCR) e aprendizado de maquina, para realizar o processo de reconhecimento dos caracteres em uma placa de identificacao de veiculo a partir de uma imagem. Como este processo esta sob condicoes reais, existem dificuldades no tocante aos diversos tipos de ruidos que podem estar presentes na aquisicao das imagens, tais como iluminacao nao uniforme, oclusao de caracteres devido a algum tipo de objeto fixado a placa de identificacao do veiculo e posicionamento dos caracteres, entre outros. Dado que o metodo de segmentacao aplicado tem influencia direta na atenuacao do efeito negativo destes ruidos, um dos objetivos deste trabalho foi determinar qual destes metodos otimiza o desempenho tanto do OCR desenvolvido quanto do modulo de OCR do mecanismo Tesseract um software considerado estado da arte nesta area. Por fim, utilizando a precisao e a revocacao como metodos de avaliacao, analisou-se qual tecnica de aprendizado de maquina obtem os melhores resultados no reconhecimento de caracteres e comparou-se estes resultados com o Tesseract, a fim de verificar-se o desempenho da abordagem proposta.

Teoria espaço-escala: conceitos e abordagens

Abordagens multi-escala vem sendo amplamente utilizadas em diversas aplicacoes de analise e processamento de sinais, sendo fundamentais em casos onde nao existem informacoes preliminares sobre a escala de observacao apropriada. A ideia basica consiste em criar uma familia de sinais derivados, permitindo assim a analise de diferentes niveis de representacao para escolha daqueles que exibem as caracteristicas de interesse. A teoria espaco-escala e uma destas abordagens. A partir dela, podem ser estabelecidas as condicoes necessarias para a definicao de transformacoes que possibilitem a manipulacao de caracteristicas presentes em diferentes niveis de maneira consistente. Este trabalho traz uma breve revisao das principais abordagens espaco-escala, bem como de suas propriedades fundamentais.

Multiscale Methods for Image Processing: The Wavelet and the Scale-Space Approaches

Multiscale approaches have been largely considered in several signal processing applications. They play an important role when designing automatic methods to cope with real world measurements where, in most of the cases, there is no prior information about which would be the appropriate scale. The basic idea behind a multiscale analysis is to embed the original signal into a family of derived signals, thus allowing the analysis of different representation levels and, further, the choice of the ones exhibiting the interest features. This paper presents a brief survey of two broadly used multiscale formulations, namely, wavelets and scale-space filtering. We present the basic definitions and some possible applications of these approaches in image processing.