Automated pollen recognition, is a problematic studied by researchers since 1996, because, palynology is a labor intensive task and there are many exciting applications to palynology such as chronological dating, climatology, allergy treatment, and even honey characterization. Many studies using different kinds of methods (chemical, optical microscopy, electronic microscopy) have been done, but the most used way as for now consist of capturing pollen slice images with an optical microscope and analyze them with image processing and machine learning tools because this procedure is cheap and accessible. So, our work is based on optical microscopy, morphological and textural attributes, and logiboost algorithm. This method has been tested on the eight different Caribbean pollens species. Firstly, pollen slices have been prepared by a palynologist. Images of these slices had been obtained with Leica DM2500 optical microscope. The pollen present in the picture had been extracted by an image processing procedure. The RGB image is converted to HSV image then Sobel operator is used, finally Otsu threshold is applied. The resulting image is blurred with Gaussian blur and thresholded again with Otsu. With an algorithm of contour detection, the holes present inside the regions are removed. Finally, the mask obtained is applied to the original image to extract the pollens. The following features had been extracted from the obtained pollen: Area Number of pixels representing pollen size and Perimeter Number of pixel of the pollen boundary. The attribute extracted at the previous stage had been used with logiboost classifier in order to determinate the species of pollen samples. To recognize the pollen species, the classifier had been trained with labeled pollen examples. The training process is as follow: Many little trees known as decision is generated and each of them is trained with the samples. By following, the success and the failure classification for each example, logiboost attribute a coefficient of trustability to each decision stump. So the response of a decision stump with a high coefficient is more valuable. Finally, the final class chosen for the submitted example is the ponderate sum of each decision stump response. To conclude, we have proposed a complete automated method to classify pollen species. This method is based on the usage of traditional image processing attribute with logiboost classifier. We obtain 93, 711 % success rate. This result is encouraging, and further study will be dedicated to the application of this procedure to a higher number of species and to honey characterization.