Finding Facial Emotions From the Clutter Scenes Using Zernike Moments-Based Convolutional Neural Networks

Face Detection

Finding human face in a cluttered scene is difficult and is also of prime importance in face-related applications. Face detection could be considered as a specific case of object detection. Object detection aims at finding the locations and dimensions of all the objects present in an image and categorize them into different classes. Cluttered images contain not only human faces but also other objects and background as displayed in Fig.1. Thus given a cluttered image, the task of the face detection system is to detect, locate the facial regions and separate or segment them in the cluttered images ignoring the other objects.

Figure 1.

Cluttered image with human faces and background

It is an undeniable fact that human faces appear with a different orientation, head poses and scales, making face detection a demanding task.

Additionally, the illumination conditions, facial expressions variation, the nonrigidity of faces, and the occlusions of objects such as glasses, beards, mustache scarves, hats and one person hiding the other add considerably to the variations in appearance of faces in an image. These added conditions increase the variability of the face patterns that face detection system should handle. Accordingly, these parameters should be taken into consideration while designing a face detection system. To get the best results it requires good algorithms and design. In recent years, there is a large increase in the number and variety of methods attributed to face detection.

A review of the related works find several methods proposed and presented on face detection using appearance-based, template matching and structural methods. The basic face detection algorithms focused on the detection of frontal faces whereas newer algorithms attempt to solve the more general and difficult problem of multiview face detection.