Sébastien Chabrier

optimization-based image segmentation by genetic algorithms

Many works in the literature focus on the definition of evaluation metrics and criteria that enable to quantify the performance of an image processing algorithm. These evaluation criteria can be used to define new image processing algorithms by optimizing them. In this paper, we propose a general scheme to segment images by a genetic algorithm. The developed method uses an evaluation criterion which quantifies the quality of an image segmentation result. The proposed segmentation method can integrate a local ground truth when it is available in order to set the desired level of precision of the final result. A genetic algorithm is then used in order to determine the best combination of information extracted by the selected criterion. Then, we show that this approach can either be applied for gray-levels or multicomponents images in a supervised context or in an unsupervised one. Last, we show the efficiency of the proposed method through some experimental results on several gray-levels and multicomponents images.

Comparative study of contour detection evaluation criteria based on dissimilarity measures

We present in this article a comparative study of well-known supervised evaluation criteria that enable the quantification of the quality of contour detection algorithms. The tested criteria are often used or combined in the literature to create new ones. Though these criteria are classical ones, none comparison has been made, on a large amount of data, to understand their relative behaviors. The objective of this article is to overcome this lack using large test databases both in a synthetic and a real context allowing a comparison in various situations and application fields and consequently to start a general comparison which could be extended by any person interested in this topic. After a review of the most common criteria used for the quantification of the quality of contour detection algorithms, their respective performances are presented using synthetic segmentation results in order to show their performance relevance face to undersegmentation, oversegmentation, or situations combining these two perturbations. These criteria are then tested on natural images in order to process the diversity of the possible encountered situations. The used databases and the following study can constitute the ground works for any researcher who wants to confront a new criterion face to well-known ones.

Hue and Saturation in the RGB Color Space

While the RGB color model refers to the biological processing of colors in the human visual system, the HSV color model corresponds to the human perception of color similarity. In this paper we formulate a projection of RGB vectors within the RGB color space, which separates achromatic from chromatic information. The projection is the mathematical equivalent to Hue and Saturation of the HSV color space in the RGB space. It integrates the psycho- visual concept of human differentiation between colors of the HSV space into the physiological-visual based concept of the RGB space. With the projection it is, contrary to the prevailing opinion, possible to differentiate between colors based on human perception in the linear geometry of the RGB color space. This opens new possibilities in many fields of color image processing, especially in the domain of color image segmentation, where color similarity plays a major role.

Ground truth method assessment for SVM-based landscape classification

Researches on land cover classification have a complete lack of ground truth methodology description. We propose a method to track ecotones as privileged training areas for SVM-based natural vegetation classification. This guidance method combines (i) the construction of a principal component analysis (PCA) on spectral bands and gray level co-occurence matrix texture attributes calculated on very high resolution images and (ii) the use of the Sobels edge detection algorithm on this PCA. The experiment is successfully applied with an overall accuracy of 82 %. Using SVM, a minimum number of mixed pixels is necessary but they can help significantly in locating an appropriate hyperplane. Moreover, the presented results show that the training stage could be more influential on classifier accuracy than classifiers themselves.

Yes, it turns: experimental evidence of pearl rotation during its formation

Cultured pearls are human creations formed by inserting a nucleus and a small piece of mantle tissue into a living shelled mollusc, usually a pearl oyster. Although many pearl observations intuitively suggest a possible rotation of the nucleated pearl inside the oyster, no experimental demonstration of such a movement has ever been done. This can be explained by the difficulty of observation of such a phenomenon in the tissues of a living animal. To investigate this question of pearl rotation, a magnetometer system was specifically engineered to register magnetic field variations with magnetic sensors from movements of a magnetic nucleus inserted in the pearl oyster. We demonstrated that a continuous movement of the nucleus inside the oyster starts after a minimum of 40 days post-grafting and continues until the pearl harvest. We measured a mean angular speed of 1.27° min−1 calculated for four different oysters. Rotation variability was observed among oysters and may be correlated to pearl shape and defects. Natures ability to generate so amazingly complex structures like a pearl has delivered one of its secrets.

Toward a Global Tuamotu Archipelago Coconut Trees Sensing Using High Resolution Optical Data

This study is part of a regeneration program of the coconut grove of French Polynesia where most coconut palm trees of the Tuamotu archipelago were planted in the 1980s following the various hurricanes that had struck islands. The French Polynesia government acquired one-meter pansharpened RGB Ikonos images over the Tuamotu archipelago. To exploit these data, a pilot study is conducted on the island of Tikehau, well-known from the specialists and easily accessible from Tahiti. A maximum likelihood (ML) classification is performed to segment the high vegetation in images. Thus, a support vector machines (SVM) classification allows the high vegetation to be classified in different patterns. And finally, a robust segmentation process based on markers controlled watershed segmentation is proposed to extract tree crowns. Through the ground mission, the trees detection accuracy is estimated which is then used to compute the number of trees the closest to the reality by applying a weighted factor to the number of trees located in each class.

Automatic Nacre Thickness Measurement of Tahitian Pearls

In this paper a methodology for an automatized measurement of the nacre thickness of Tahitian pearls is presented. An adapted snake approach as well as our own developed circle detection algorithm are implemented to extract the nacre boundaries out of X-ray images. The results are validated by experts currently performing manually the obligatory nacre thickness control for millions of Tahitian pearls that are exported each year. Equivalent articles propose methods suitable for round pearls, whereas this paper contains methods to evaluate the nacre profile of pearls independently of their shape. As the algorithms are not specifically parametrized for Tahitian pearls, the methods can be adapted for quality assessment of other pearls as well.

Automatic Classification of Tahitian Pearls

Tahitian pearls are currently classified manually by pearl experts. This process is time consuming and furthermore subjective for some of the quality parameters such as colour, form or lustre. In this paper we present our ongoing work to implement an automatic classification of Tahitian pearls out of images. For this purpose different image segmentation and machine learning methods are used. In the following sections we explain our methodology, show first results for several sub-steps and give new ideas for greylevel edge detection and colour classification.

SVM texture classification for tropical vegetation mapping

Nowadays, remote sensing is an essential science in French Polynesia because of its extended territory and the remoteness of its 120 islands. There is a strong need to study the vegetation cover and its evolution (biodiversity threat, invasive species, etc.). A growing satellite images database has been acquired throughout, giving access to very high resolution optical images such as Quickbird data. These data allow accessing the vegetation canopy spectral and contextual information, texture classification has proved to be an efficient tool to map the complex vegetation found in tropical regions. The main goal of this paper is to propose an optimized SVM multispectral-texture classification method for tropical vegetation mapping. One of the texture computation drawbacks is the window treatment size, which is related to the largest texture element size. In complex tropical vegetation cover, this parameter leads to very small ground truth learning database, inducing a significant degradation of the classifications accuracy. We propose to increase the thumbnail numbers using an under-sampling method, optimizing the size and the number of the thumbnails. The other drawback is the high dimensionality of the problem when dealing with multispectral textures. We thus propose to rank and select the most pertinent textures attributes in order to reduce the dimensionality without reducing the classification accuracy. We first introduce the study context, before exposing preliminary studies on tuning the SVM learning method. The adapted method is then accurately exposed and the interesting experimental results as well as a sample of applications are presented before to conclude.

Multi-source SVM fusion for environmental monitoring in Marquesas archipelago

Mapping plant species in montane tropical ecosystems needs the use of complementary information sources to be optimally accurate. In this paper, we study SVM fusion as a tool to classify several sources as optical, synthetic aperture radar and topographical ones. Our fusion scheme consists first in applying a single SVM on each individual data. Their outputs are then used for a SVM-based decision fusion to predict the final class membership of each sample. SVM fusion outperforms all mono-source SVM, our fusion method showing numerous successful traits.