Mal_CNN: An Enhancement for Malicious Image Classification Based on Neural Network

Abstract

In image analysis and processing, image segmentation is one of the most important functions. The outcomes of segmentation have such significance on all subsequent image analysis operations, covering object tracking and description, feature measurement, and even higher-level tasks like object recognition. Malicious code detection is becoming increasingly significant, and current models must be improved. Hence forth the Image segmentation in the field of Malware image classification is a significant task. The sectional structure or region of interest must be identified and extracted during the segmentation process so that it can be evaluated independently. There are various reviews stating the traditional approach of image segmentation in various fields. The necessity of image segmentation in malicious image is extracting data for classification using CNN is discussed. In this work we use malimg_paper_dataset_imgs with 9,339 malware images. Various segmentation techniques were used to enhance the malware images. Those enhanced malicious image were applied in CNN architecture and Mal_CNN for classification and a comparative result is been discussed. The malicious images in dataset after incorporating segmentation have achieved 95% of accuracy in CNN architecture and 97% with Mal_CNN.

Keywords:

• Edge
• histogram
• object recognition
• segmentation
• threshold