Achieving human-level interpretation of visual complexity will have numerous applications in data hiding, image compression, image retrieval, computer vision, etc. Previous studies relied on using unsupervised learning to coalesce handcrafted image features, such as edges and colours, for assessment of visual complexity. Our study utlises the potency of Convolutional Neural Networks (CNNs) to improve the classification accuracy and assessment of visual complexity based on the Corel 1000A dataset. We incorporated SVM-based supervised learning to classify the features extracted by the CNN. Furthermore, we exploited the utility offered by fine tuning and appropriate adjustments to the CNN structure that were incorporated into our learning strategy which led to 13.6 % improvement in classification accuracy than the available (unsupervised) and supervised learning methods.