Expoente direcional de Hurst na análise de similaridade de imagens de sementes

Abstract

Modernization is present in all fields of knowledge. Increasingly sophisticated techniques and more modern devices have come up frequently. Wavelet decomposition is a tool that has fundamental importance in many of these advances. With regard to image analysis, this tool has cooperated to create several new techniques, like the ones for reconstruction, compression, noise elimination, among others. Another tool that assists in the analysis of images is the Hurst exponent, which measures how self-similar an image is, so that information about the characteristics of the image is captured, which would be impossible with the naked eye. Therefore, the objective of this work will be to combine the technique of wavelet decomposition with the calculation of the Hurst exponent to analyze seed images and thus be able to classify them as full, slightly damaged or damaged. In the calculation of the Hurst exponent it will be used as location measure the mean and the median. A suport vector machine will be used for validation of the proposed method. For the group of all seeds the mean accuracy of the method, using the mean, was 74.5 % and with the median was 57.05 %. Using the group of full and damaged seeds the mean accuracy rate, with the mean as a measure of position, was 99.76 % and with the median was 80.93 %. In the group containing slightly damaged and damaged seeds the mean accuracy rate, using the mean as a measure of position, was 99.26 % and with the median was 76.22 %.