A FRAMEWORK FOR MORPHOLOGICAL OPERATIONS USING COUNTER HARMONIC MEAN

Abstract

In this article, we have a tendency to embrace a novel framework for learning morphological operations using counter-harmonic mean. It combines the conception of morphology with convolutional neural networks. Similarly, the elemental morphological operators of dilation and erosion, opening and closing, as well as the more refined top-hat transform, for which we disclose a real-world application from the steel industry, are all subjected to a rigorous experimental validation. Our system learns about the structuring element and the operator's composition via online learning and stochastic gradient descent. It works effectively with massive datasets and in online environments.

Keywords

• mathematical morphology
• convolutional neural networks
• on-line mastering and strutting element