Population-based continuous optimization, probabilistic modelling and mean shift

Evolutionary algorithms perform optimization using a population of sample solution points. An interesting development has been to view population-based optimization as the process of evolving an explicit, probabilistic model of the search space. This paper investigates a formal basis for continuous, population-based optimization in terms of a stochastic gradient descent on the Kullback-Leibler divergence between the model probability density and the objective function, represented as an unknown density of assumed form. This leads to an update rule that is related and compared with previous theoretical work, a continuous version of the population-based incremental learning algorithm, and the generalized mean shift clustering framework. Experimental results are presented that demonstrate the dynamics of the new algorithm on a set of simple test problems.

Can mean shift trackers perform better?

Many tracking algorithms have difficulties dealing with occlusions and background clutters, and consequently don't converge to an appropriate solution. Tracking based on the mean shift algorithm has shown robust performance in many circumstances but still fails e.g. when encountering dramatic intensity or colour changes in a pre-defined neighbourhood. In this paper, we present a robust tracking algorithm that integrates the advantages of mean shift tracking with those of tracking local invariant features. These features are integrated into the mean shift formulation so that tracking is performed based both on mean shift and feature probability distributions, coupled with an expectation maximisation scheme. Experimental results show robust tracking performance on a series of complicated real image sequences. © 2010 IEEE.

Deteção e acompanhamento de movimento através de uma câmara de vídeo

A deteção e seguimento de pessoas tem uma grande variedade de aplicações em visão computacional. Embora tenha sido alvo de anos de investigação, continua a ser um tópico em aberto, e ainda hoje, um grande desafio a obtenção de uma abordagem que inclua simultaneamente exibilidade e precisão. O trabalho apresentado nesta dissertação desenvolve um caso de estudo sobre deteção e seguimento automático de faces humanas, em ambiente de sala de reuniões, concretizado num sistema flexível de baixo custo. O sistema proposto é baseado no sistema operativo GNU's Not Unix (GNU) linux, e é dividido em quatro etapas, a aquisição de vídeo, a deteção da face, o tracking e reorientação da posição da câmara. A aquisição consiste na captura de frames de vídeo das três câmaras Internet Protocol (IP) Sony SNC-RZ25P, instaladas na sala, através de uma rede Local Area Network (LAN) também ele já existente. Esta etapa fornece os frames de vídeo para processamento à detecção e tracking. A deteção usa o algoritmo proposto por Viola e Jones, para a identificação de objetos, baseando-se nas suas principais características, que permite efetuar a deteção de qualquer tipo de objeto (neste caso faces humanas) de uma forma genérica e em tempo real. As saídas da deteção, quando é identificado com sucesso uma face, são as coordenadas do posicionamento da face, no frame de vídeo. As coordenadas da face detetada são usadas pelo algoritmo de tracking, para a partir desse ponto seguir a face pelos frames de vídeo subsequentes. A etapa de tracking implementa o algoritmo Continuously Adaptive Mean-SHIFT (Camshift) que baseia o seu funcionamento na pesquisa num mapa de densidade de probabilidade, do seu valor máximo, através de iterações sucessivas. O retorno do algoritmo são as coordenadas da posição e orientação da face. Estas coordenadas permitem orientar o posicionamento da câmara de forma que a face esteja sempre o mais próximo possível do centro do campo de visão da câmara. Os resultados obtidos mostraram que o sistema de tracking proposto é capaz de reconhecer e seguir faces em movimento em sequências de frames de vídeo, mostrando adequabilidade para aplicação de monotorização em tempo real.

Empresa Davantis: seguimiento de personas y vehículos por videovigilancia inteligente

Este proyecto, desarrollado en la empresa Davantis, tiene como objetivo encontrar posibles mejoras a su actual sistema de videovigilancia, el Daview. El proyecto está dedicado al estudio del algoritmo de seguimiento Mean Shift para la elaboración de un sistema de tracking. Para ello se han desarrollado y evaluado tres implementaciones diferentes, mediante las cuales se han encontrado mejoras que complementan al módulo de tracking del Daview. También se ha estudiado la utilidad de un sistema de evaluación manual frente a uno de automático.

Developing raster bands computed from LiDAR to improve forest

This study examines the use of di erent features derived from remotely sensed data in segmentation of forest stands. Surface interpolation methods were applied to LiDAR points in order to represent data in the form of grayscale images. Median and mean shift ltering was applied to the data for noise reduction. The ability of di erent compositions of rasters obtained from LiDAR data and an aerial image to maximize stand homogeneity in the segmentation was evaluated. The quality of forest stand delineations was assessed by the Akaike information criterion. The research was performed in co-operation with Arbonaut Ltd., Joensuu, Finland.

Preliminary results concerning the VSS X- chart with unknow in-control parameters

The VSS X- chart is known to perform better than the traditional X- control chart in detecting small to moderate mean shifts in the process. Many researchers have used this chart in order to detect a process mean shift under the assumption of known parameters. However, in practice, the process parameters are rarely known and are usually estimated from an in-control Phase I data set. In this paper, we evaluate the (run length) performances of the VSS X- control chart when the process parameters are estimated and we compare them in the case where the process parameters are assumed known. We draw the conclusion that these performances are quite different when the shift and the number of samples used during the phase I are small. ©2010 IEEE.

Land use image classification through optimum-path forest clustering

Land use classification has been paramount in the last years, since we can identify illegal land use and also to monitor deforesting areas. Although one can find several research works in the literature that address this problem, we propose here the land use recognition by means of Optimum-Path Forest Clustering (OPF), which has never been applied to this context up to date. Experiments among Optimum-Path Forest, Mean Shift and K-Means demonstrated the robustness of OPF for automatic land use classification of images obtained by CBERS-2B and Ikonos-2 satellites. © 2011 IEEE.

Toward Satellite-Based Land Cover Classification Through Optimum-Path Forest

Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP)

Applying enhancement filters in the pre-processing of images of lymphoma

Lymphoma is a type of cancer that affects the immune system, and is classified as Hodgkin or non-Hodgkin. It is one of the ten types of cancer that are the most common on earth. Among all malignant neoplasms diagnosed in the world, lymphoma ranges from three to four percent of them. Our work presents a study of some filters devoted to enhancing images of lymphoma at the pre-processing step. Here the enhancement is useful for removing noise from the digital images. We have analysed the noise caused by different sources like room vibration, scraps and defocusing, and in the following classes of lymphoma: follicular, mantle cell and B-cell chronic lymphocytic leukemia. The filters Gaussian, Median and Mean-Shift were applied to different colour models (RGB, Lab and HSV). Afterwards, we performed a quantitative analysis of the images by means of the Structural Similarity Index. This was done in order to evaluate the similarity between the images. In all cases we have obtained a certainty of at least 75%, which rises to 99% if one considers only HSV. Namely, we have concluded that HSV is an important choice of colour model at pre-processing histological images of lymphoma, because in this case the resulting image will get the best enhancement.

Segmentation and Recognition of Multi-Food Meal Images for Carbohydrate Counting

In this paper, we propose novel methodologies for the automatic segmentation and recognition of multi-food images. The proposed methods implement the first modules of a carbohydrate counting and insulin advisory system for type 1 diabetic patients. Initially the plate is segmented using pyramidal mean-shift filtering and a region growing algorithm. Then each of the resulted segments is described by both color and texture features and classified by a support vector machine into one of six different major food classes. Finally, a modified version of the Huang and Dom evaluation index was proposed, addressing the particular needs of the food segmentation problem. The experimental results prove the effectiveness of the proposed method achieving a segmentation accuracy of 88.5% and recognition rate equal to 87%

Graphical Model-Based Vertebra Identification from X-Ray Image(s)

Automated identification of vertebrae from X-ray image(s) is an important step for various medical image computing tasks such as 2D/3D rigid and non-rigid registration. In this chapter we present a graphical model-based solution for automated vertebra identification from X-ray image(s). Our solution does not ask for a training process using training data and has the capability to automatically determine the number of vertebrae visible in the image(s). This is achieved by combining a graphical model-based maximum a posterior probability (MAP) estimate with a mean-shift based clustering. Experiments conducted on simulated X-ray images as well as on a low-dose low quality X-ray spinal image of a scoliotic patient verified its performance.

Generalized methods and solvers for noise removal from piecewise constant signals. I. Background theory

Removing noise from piecewise constant (PWC) signals is a challenging signal processing problem arising in many practical contexts. For example, in exploration geosciences, noisy drill hole records need to be separated into stratigraphic zones, and in biophysics, jumps between molecular dwell states have to be extracted from noisy fluorescence microscopy signals. Many PWC denoising methods exist, including total variation regularization, mean shift clustering, stepwise jump placement, running medians, convex clustering shrinkage and bilateral filtering; conventional linear signal processing methods are fundamentally unsuited. This paper (part I, the first of two) shows that most of these methods are associated with a special case of a generalized functional, minimized to achieve PWC denoising. The minimizer can be obtained by diverse solver algorithms, including stepwise jump placement, convex programming, finite differences, iterated running medians, least angle regression, regularization path following and coordinate descent. In the second paper, part II, we introduce novel PWC denoising methods, and comparisons between these methods performed on synthetic and real signals, showing that the new understanding of the problem gained in part I leads to new methods that have a useful role to play.

Generalized methods and solvers for noise removal from piecewise constant signals. II. New methods

Removing noise from signals which are piecewise constant (PWC) is a challenging signal processing problem that arises in many practical scientific and engineering contexts. In the first paper (part I) of this series of two, we presented background theory building on results from the image processing community to show that the majority of these algorithms, and more proposed in the wider literature, are each associated with a special case of a generalized functional, that, when minimized, solves the PWC denoising problem. It shows how the minimizer can be obtained by a range of computational solver algorithms. In this second paper (part II), using this understanding developed in part I, we introduce several novel PWC denoising methods, which, for example, combine the global behaviour of mean shift clustering with the local smoothing of total variation diffusion, and show example solver algorithms for these new methods. Comparisons between these methods are performed on synthetic and real signals, revealing that our new methods have a useful role to play. Finally, overlaps between the generalized methods of these two papers and others such as wavelet shrinkage, hidden Markov models, and piecewise smooth filtering are touched on.

Fuzzy c-means variants for medical image segmentation

Segmentation is an important step in many medical imaging applications and a variety of image segmentation techniques exist. One group of segmentation algorithms is based on clustering concepts. In this article we investigate several fuzzy c-means based clustering algorithms and their application to medical image segmentation. In particular we evaluate the conventional hard c-means (HCM) and fuzzy c-means (FCM) approaches as well as three computationally more efficient derivatives of fuzzy c-means: fast FCM with random sampling, fast generalised FCM, and a new anisotropic mean shift based FCM. © 2010 by IJTS, ISDER.

Color-based surface reflectance separation for scene illumination estimation and rendering

Given the importance of color processing in computer vision and computer graphics, estimating and rendering illumination spectral reflectance of image scenes is important to advance the capability of a large class of applications such as scene reconstruction, rendering, surface segmentation, object recognition, and reflectance estimation. Consequently, this dissertation proposes effective methods for reflection components separation and rendering in single scene images. Based on the dichromatic reflectance model, a novel decomposition technique, named the Mean-Shift Decomposition (MSD) method, is introduced to separate the specular from diffuse reflectance components. This technique provides a direct access to surface shape information through diffuse shading pixel isolation. More importantly, this process does not require any local color segmentation process, which differs from the traditional methods that operate by aggregating color information along each image plane. ^ Exploiting the merits of the MSD method, a scene illumination rendering technique is designed to estimate the relative contributing specular reflectance attributes of a scene image. The image feature subset targeted provides a direct access to the surface illumination information, while a newly introduced efficient rendering method reshapes the dynamic range distribution of the specular reflectance components over each image color channel. This image enhancement technique renders the scene illumination reflection effectively without altering the scene’s surface diffuse attributes contributing to realistic rendering effects. ^ As an ancillary contribution, an effective color constancy algorithm based on the dichromatic reflectance model was also developed. This algorithm selects image highlights in order to extract the prominent surface reflectance that reproduces the exact illumination chromaticity. This evaluation is presented using a novel voting scheme technique based on histogram analysis. ^ In each of the three main contributions, empirical evaluations were performed on synthetic and real-world image scenes taken from three different color image datasets. The experimental results show over 90% accuracy in illumination estimation contributing to near real world illumination rendering effects. ^