Multi View Face Video Super Resolution

The most significant issues in recognition of face are computer vision and pattern identification where high-resolution spatial face images or videos required improving the pictographic information for human analysis and automatic machine observation depiction. Image resolution depends upon the pixel size, and it explains the details present in an image. Super Resolution (SR) has several applications such as surveillance of video, intelligent ID card and access control.

The normal illumination, HR imaging and frontal view in the controlled environment conditions, the existing face identification algorithms may obtain high identification rates. The recognition of face is a challenging task in which only facial images of low resolution (LR) are accessible with high pose variations. There is a need to deal with "one sample per class problem", in which the HR images of having one frontal are present in passports, personal certificates or ID cards, etc. in the gallery. In video surveillance, the high resolution image of frontal face recognizes the face of an individual person, generally aims at identifying the low resolution of images of non-frontal face (due to large distance between camera and object) from the gallery. In-addition there are various problems in the traditional identification of face techniques, such as variation in poses; each person has only one image in the gallery and resolution differences in image etc. Hence, in real-world applications, face identification is still a difficult task while only facial images of LR are offered through large poses variations.

Another important aspect which emphasizes the use of SR algorithms is that most widely used imaging sensors like Complementary Metal Oxide Semiconductor (CMOS) and Charge-Coupled Devices (CCDs) arrays, also have a limitation on the spatial resolution of the sensors. Hardware technology and the manufacturing of optics technology are not able to support the order of expected resolution also created a need to learn SR algorithms to attain the resolution improvement goal.