A novel approach using deep convolutional neural network to classify the photographs based on leading-line by fine-tuning the pre-trained VGG16 neural network

Leading lines are used to draw the viewer's attention towards the main subject present in a photograph. These lines may be straight or curved, and generally, start from the left or right bottom corner of the frame. It leads the viewer's eyes to the region of interest, which could be any object or the vanishing point. It has been observed that the photograph containing leading lines achieves a higher aesthetic score in several photographic competitions and different photo-sharing communities. Automatic detection of the leading line may solve numerous real-time applications like on-site aesthetic value evaluation, assisting amateur photographer to get a better composition, and so on. However, implementation of the same is not easy due to the subjective nature of the problem. From the literature survey, we can notice that no such work exists that considers the presence of leading line in a photograph. This paper introduces a novel approach using deep convolutional neural network (DCNN) framework to detect the presence of leading line in a photograph as well as classify the photographs based on the leading line. The proposed model automatically extracts features by fine-tuning the pre-trained VGG16 CNN (a transfer learning method) to accomplish the classification task. The model has been examined by executing on the ground truth dataset, and the satisfying result has been observed. Another contribution of the proposed work is creating a new dataset named as Leading-line Containing Image (LCI) which will be beneficial for future work in this area.