Hybrid Feature Learning and Engineering Based Approach for Face Shape Classification

Face shape classification is a vital process to choose an appropriate eyelashes, hairstyle and facial makeup, and selection of a suitable glasses' frames according to the guidelines from experts. Measuring face characteristics by beauty experts manually costs time and efforts. Therefore, developing automated face shape identification system could alleviate the need for additional time and efforts made by experts. Many automatic face shape classification methods have been proposed in the literature; however, the existing methods tackle many challenges due to the complexity of face geometry and variation in its characteristics. This paper presents a deep convolutional neural network method (CNN) approach for classifying face shape into five types. The proposed method which is based on merging the features learnt by CNN with hand crafted features represented by histogram of oriented gradients (HOG) and facial landmarks has proven to be efficient in identification of facial shape. The obtained results demonstrate that the proposed method is promising in identifying the shape of face achieving accuracy of 81.1%.