Segmentation-based Euler number with multi-levels for image feature description

This paper proposes a new and efficient image feature descriptor using Euler Number with the help of segmentation according to given number of levelsest. The proposed Segmentation-based Euler Number for image description algorithm (SENA) works as the following steps. First, transforming the image into gray image if the image is color image; then, dividing the gray image into different sets using the given number of levelsets; next, decomposing the gray image into binary images with multi-thresh using the otsu algorithm; following, computing the Euler Number of each binary image; finally, combining the Euler Numbers, mean and variance to form the feature vector for an input image. The proposed SENA was employed to the image classification on three public available dataset (Stanford Dogs Dataset, 17 flower dataset, and Caltech 256 dataset). We compute SENA with LBP and Gabor on the Stanford Dogs Dataset, the detail classification results on 17 flower dataset is given as confusion matrix, and the result of SENA on the Caltech 256 dataset is compared with those of the recently reported. The experiments demonstrate a competitive performance of SENA for classification task.