Fog Classification and Accuracy Measurement Using SVM

Fog is not always homogeneous in nature. The fog density and distribution are varying in nature while capturing images through a camera or sensor. In contrast to homogeneity the fog may be treated as heterogeneous which depends upon the density variation of its constituents particles i.e water droplets. Classification is important and sometimes helpful to design a fog removal algorithm for vision enhancement while considering type of fog without knowing its density. Classification methods are applicable for both synthetic and camera images. This paper presents Support Vector Machine (SVM) that plays a key role to classify the synthetic data into two classes with accuracy measurement. Confusion matrix and Receiver Operational Characteristic (ROC) curve hold SVM to quantify the accuracy. The proposed method quantifies the type of fog with more than 92 percent accuracy for synthetically generated images containing various objects and environments in foggy situation. This acquaintance will finally help to generate a natural image dataset of homogeneous and heterogeneous foggy images.