Morphological Operations and Artificial Neural Networks for Multi-scale colored texture classification

Among image descriptors, texture is one of the most used features to represent the visual content of images. According to several studies, texture could be seen differently based on the selected scale of interest (SoI), which raises a serious problem of finding the appropriate SoI. To tackle this issue, a new scheme based on mathematical morphology and artificial neural networks (ANN) has been introduced. Firstly, the image is subjected to a series of morphological operations to extract the different SoIs. From each SoI, a set of color features are extracted and fed to two successive ANNs in order to categorize the textures. The present scheme maintains the extraction of both color and multi-scale texture information. A comprehensive experimental comparison has been conducted where the present scheme has outperformed other widely known texture and CNN-based descriptors.