Automatic Date Fruit Classification by Using Local Texture Descriptors and Shape-Size Features

Abstract

In this paper, we propose a system of automatically classifying different types of dates from their images. Different dates have various distinguished features that can be useful to recognize a particular date. These features include color, texture, and shape. In the proposed system, a color image of a date is decomposed into its color components. Then, local texture descriptor in the form of local binary pattern (LBP) or Weber local descriptor (WLD) histogram is applied to each of the components to encode the texture pattern of the date. The texture patterns from all the components are fused to describe the image. Fisher discrimination ratio (FDR) based feature selection is utilized to reduce the dimensionality of the feature set. Size and shape features are appended to the texture descriptors to fully describe the date. As a classifier, we use support vector machines. The proposed system achieves more than 99% accuracy to classify the dates and outperforms previous method of dates classification.