Topics in Medical Image Processing and Computational Vision

Object detection is challenging when the object class exhibits large within-class variations. In this work, we show that foreground-background classification (detection) and within-class classification of the foreground class (pose estimation) can be jointly learned in a multiplicative form of two kernel functions. Model training is accomplished via standard SVM learning. When the foreground object masks are provided in training, the detectors can also produce object segmentations. A tracking-by-detection framework to recover foreground state in video sequences is also proposed with our model. The advantages of our method are demonstrated on tasks of object detection, view angle estimation and tracking. Our approach compares favorably to existing methods on hand and vehicle detection tasks. Quantitative tracking results are given on sequences of moving vehicles and human faces.

A 3D modular morphable model (3DMMM) is introduced to deal with facial expression recognition. The 3D Morphable Model (3DMM) contains 3D shape and 2D texture information of faces extracted using conventional Principal Component Analysis (PCA). In this work, modular PCA approach is used. A face is divided into six modules according to different facial features which are categorized based on Facial Animation Parameters (FAP). Each region will be treated separately in the PCA analysis. Our work is about recognizing the six basic facial expressions, provided that the properties of a facial expression are satisfied. Given a 2D image of a subject with facial expression, a matched 3D model for the image is found by fitting them to our 3D MMM. The fitting is done according to the modules; it will be in order of the importance modules in facial expression recognition (FER). Each module is assigned a weighting factor based on their position in priority list. The modules are combined and we can recognize the facial expression by measuring the similarity (mean square error) between input image and the reconstructed 3D face model.

With the Internet evolution, there has been a huge increase in the amount of images stored in electronic format particularly in the case of biological and medical image applications. Nowadays, hospitals and research centers can acquire large image databases which poses fundamental requirements for storage, processing and sharing data. In order to fulfill these requirements we have proposed the Shared Biological Image Manager (SBIM) system which has been developed using the programming languages PHP and Javascript as well as the Database Management System PostgreSQL. In this chapter, we propose an extension of the SBIM functionalities by incorporating data mining and image retrieval facilities. We describe an unified solution for both these services inside the Shared Biological Image Manager (SBIM) through Support Vector Machine (SVM) frameworks. Data mining is implemented using discriminant weights given by SVM separating hyperplanes to select the most discriminant features for two-class classification problems. For image retrieval, we consider an SVM ensemble based on the “one-against-all” SVM multi-class approach. The user specifies an initial feature space, the training set and the SVM (ensemble) configuration. After training, the SVM ensemble can be used to retrieve relevant data once given a query image. Finally, we discuss some details about the implementation of the content-based image retrieval (CBIR) and discriminant features discovery approaches inside the SBIM system.

The manifestation of pathogens in plantations is the most important cause of losses in several crops. These usually represent less income to the farmers due to the lower product quality as well as higher prices to the consumer due to the smaller offering of goods. The sooner the disease is identified the sooner one can control it through the use of agrochemicals, avoiding great damages to the plantation. This chapter introduces a method for the automatic classification of cotton diseases based on the feature extraction of foliar symptoms from digital images. The method uses the energy of the wavelet transform for feature extraction and a Support Vector Machine for the actual classification. Five possible diagnostics are provided: (1) healthy (SA), (2) injured with Ramularia disease (RA), (3) infected with Bacterial Blight (BA), (4) infected with Ascochyta Blight (AS), or (5) possibly infected with an unknown disease.

The increasing ubiquitousness of digital information in our daily lives has positioned video as a favored information vehicle, and given rise to an astonishing generation of social media and surveillance footage. This raises a series of technological demands for automatic video understanding and management, which together with the compromising attentional limitations of human operators, have motivated the research community to guide its steps towards a better attainment of such capabilities. As a result, current trends on cognitive vision promise to recognize complex events and self-adapt to different environments, while managing and integrating several types of knowledge. Future directions suggest to reinforce the multi-modal fusion of information sources and the communication with end-users.

Image edge detection is one of the more fashionable topics in image processing and it is an important preprocessing step in many image processing techniques since its performance is crucial for the results obtained by subsequent higher-level processes. In this paper, an edge detection algorithm for noisy images, corrupted with salt and pepper noise, using a fuzzy morphology based on discrete t-norms is proposed. It is shown that this algorithm is robust when it is applied to different types of noisy images. The obtained results are objectively compared with other well-known morphological algorithms such as the ones based on the Łukasiewicz t-norm, representable and idempotent uninorms and the classical umbra approach. This comparison is addressed using some different objective measures for edge detection, such as Pratt’s figure of merit, the \(\rho \)-coefficient, and noise removal like the structural similarity index and the fuzzy \(DI\)-subsethood measure. The filtered results and the edge images obtained with our approach improve the values obtained by the other approaches.

Colour quantisation is very often used as an auxiliary operation in colour image processing, e.g. this operation can reduce the complexity of image segmentation process. In this chapter the results of segmentation preceded by a colour quantisation have been compared with segmentation without such preprocessing step. The choice of tools for the experiment was, for obvious reasons, limited to some colour quantisation and image segmentation methods. The colour quantisation techniques based on clustering of pixels, i.e. the classic \(k\)-\(means\) technique (KM) and new \(k\)-\(harmonic means\) technique (KHM) were considered. For image segmentation the unseeded region growing (USRG) technique has been selected from a variety of known techniques. Evaluation of the results was based on empirically defined quality function used for segmentation results. Not every method of colour quantisation, carried out as preprocessing step in the process of segmentation, leads to improved segmentation result. Therefore, our approach needs a good quantisation technique, e.g. researched segmentation technique works better for KHM quantisation technique than KM technique. This study uses different images acquired from relatively simple scenes without significant highlights and shadows. An interesting open question is what kind of colour images needs to be quantised before the segmentation. Perhaps an estimation of image segmentation difficulty will help to answer this question. The further research should be focused on establishing the conditions and parameters of additional improvement in image segmentation preceded by a colour quantisation.

Fluids, especially physiological fluids, either incompressible such as blood or compressible such as air, can flow through strongly deformed vessels in certain circumstances. In particular, veins and intrathoracic segments of the respiratory tract can collapse when they experience strong external pressure. In addition, the digestive tract launches an active peristaltism, i.e., unidirectional waves of radially symmetrical contraction and relaxation of mural smooth muscles, to propel nutrients to destination.

Image analysis is extremely important to obtain crucial information about the blood phenomena in microcirculation. The current study proposes an automatic method for segmentation and tracking red blood cells (RBCs) flowing through a 100 μm glass capillary. The original images were obtained by means of a confocal system and then processed in MatLab using the Image Processing Toolbox. The automatic measurements with the proposed automatic method are compared with a manual tracking method performed by ImageJ. The comparison of the two methods is performed using a statistical Bland–Altman analysis. The numerical results have shown a good agreement between the two methods. Additionally, no significant difference was found between the two methods and as a result the proposed automatic method is demonstrated to be a rapid and accurate way to track RBCs in microchannels.

In this chapter, a computer-assisted system aimed to assess the degree of regeneration of bone tissue from stem cells is built. We deal with phenotype and color analysis to describe a wide variety of microscopic biomedical images. Then we investigate several trained and non-parametric classifiers based on neural networks, decision trees, bayesian classifiers and association rules, whose effectiveness is analyzed to distinguish between bone and cartilage versus other existing types of tissue existing in our input biomedical images. The features selection includes texture, shape and color descriptors, among which we consider color histograms, Zernike moments and Fourier coefficients. Our study evaluates different selections for the feature vectors to compare accuracy and computational time as well as different stainings for revealing tissue properties. Overall, picrosirius reveals as the best staining and multilayer perceptron as the most effective classifier to distinguish between bone and cartilage tissue.

Our main objective is design and develop a system that assesses sudomotor function with spatial and temporal resolution through digital image processing techniques. Its evaluation has become significant in the diagnosis of several nerve diseases. The current methods to evaluate post-ganglionic sudomotor function are not very successful because they are too expensive or they do not give enough information. It will be desirable to achieve useful results with a low cost approach. In order to this, it can be used a pH indicator on the skin of patient that changes colour when it comes in contact with sweat and a digital image processing algorithm to quantify it. The sudomotor function of more than 20 patients, with a wide range of profiles, has been tested. There is a high correlation between our results and those of others kinds of sweat tests. From all of this it can conclude that it is possible to implement an evaluation system for sudomotor function using digital image processing with a low cost solution.

Digital Medical Thermal Imaging (DMTI) is a modality of medical imaging for monitoring the surface of the skin temperature. The technology evolution over the last 50 years contributed for more accuracy in the measurements and to significantly decrease the size of the equipment making them portable today. The applications of this technique in medicine are on the peripheral vascular, neurological and musculoskeletal conditions assessing and monitoring, in such areas like: cardiology, chronic diseases, dermatology, dentistry, obstetrics, occupational medicine, oncology, physiotherapy, public health, surgery and veterinary medicine. In this chapter the technique is introduced, with its historical perspective, the fundamental physics, the human physiology concepts, the equipment characterization, the existing proposals for examination protocols, the used image processing techniques, the latest developments and applications and the current limitations and challenges.

A retinal image presents three important structures in a healthy eye: optic disk, fovea and blood vessels. These are diseases associated with changes in each of these structures. Some parameters should be extracted in order to evaluate if an eye is healthy. For example, the level of imperfection of the optic disk’s circle contour is related with glaucoma. Furthermore, the proximity of the lesion in the retina to the fovea (structure responsible for the central vision) induces loss of vision. Advanced stages of diabetic retinopathy cause the formation of micro blood vessels that increase the risk of detachment of the retina or prevent light from reaching the fovea. On the other hand, the arterio-venous ratio calculated through the thickness of the central vein and artery of the retina, is a parameter extracted from the vessels segmentation. In image processing, each structure detected has special importance to detect the others, since each one can be used as a landmark to the others. Moreover, often masking the optic disk is crucial to reach good results with algorithms to detect other structures. The performance of the detection algorithms is highly related with the quality of the image and with the existence of lesions. These issues are discussed below.

Dermoscopy is a non-invasive diagnosis technique for in vivo observation of pigmented skin lesions used in dermatology. There is currently a great interest in the development of computer assisted diagnosis systems, given their great potential to this area of medicine. The standard approach in automatic dermoscopic image analysis can be divided in three stages: image segmentation, feature extraction/selection and lesion classification. In order to validate the algorithms developed for each stage, a great number of reliable images and clinical diagnosis are required. This paper presents a software tool to collect and organize dermoscopic data from hospital databases. It is suitable for clinical daily routine and simultaneously has a data structure to support the development and validation of algorithms created by the researchers to construct the computer assisted diagnosis system. This tool is composed by a database with three related but independent modules: Clinical Module, Processing Module and Statistical Module.

In this paper Synthetic Aperture Radar (SAR) images in X-band were analyzed in order to infer ground properties from data. The aim was to classify different zones in peri-urban forestries integrating information from different sources. In particular the X band is sensitive to the moisture content of the ground that can be therefore put into relation with the gray level of the image; moreover, the gray level is related to the smoothness and roughness of the ground. An integration of image segmentation and machine learning methods is studied to classify different zones of peri-urban forestries, such as trees canopies, lawns, water pounds, roads, etc., directly from the gray level signal properties. As case study the X-SAR data of a forest near Rome, the Castel Fusano area, are analyzed.

The spatial distribution and extent of pervious and impervious areas in the city are important variables for planning, mitigating, preparing and responding to potential urban flooding events. Remote sensing constitutes a valuable data source to derive land cover information required for flood risk assessment. The present paper describes the generation of a Land Cover Map for the city of Lisbon, Portugal. The data source is an IKONOS-2 pansharp image, from 2008, with a spatial resolution of 1 m, and a normalized Digital Surface Model (nDSM) from 2006. The methodology was based on the extraction of features of interest, namely: vegetation, soil and impervious surfaces. It is demonstrated that using a methodology based on Very-High Resolution (VHR) images, quick updating of detailed land cover information is possible and can be used to support decisions in a crisis situation where official maps are generally outdated.