Research On Infrared Image Normalization Under Variant Ambient Temperature

As a burgeoning biological feature recognition technology, face recognition, being straightforward, passive and non-invasive, is one of the most active domain in the biological feature recognition. The key problem of face recognition is whether the system can adapt the faces which are changed as a result of the variations in the practical applications. Face recognition system will be affected largely by the variations of the light, pose, emotional expression, toilette, legerdemain of images and so on. However, infrared face recognition, which are not easy to affected by the toilette, skin color, expression and pose, can heavily avoid the shortage of the face recognition. Infrared face recognition has been an important branch in biological feature recognition. This paper analyzes the advantage and shortage of infrared facial images and deeply studies the normalization methods of thermal images.This paper analyzes the characteristics of the infrared facial images first, summarize the factors affect the variations of infrared facial Hot-Mode, introduce the normalization methods of thermal images, illumination normalization and radiometric normalization of multitemporal satellite imagery, which the later two normalization methods have the same application background. Then three temperature normalization methods are proposed in three aspects to eliminate or reduce the effect of ambient temperature.Although infrared face recognition is not affected by the light, the effects of ambient temperature are equal to the influence of light on face recognition. Massive research has been done in this article. From the statistical aspect, the average and variance of images are used to analyze the law between the value of thermal images and the ambient temperature. We study the change of temperature relations in block-images along with the ambient temperature. Then three kinds of temperature normalization methods are proposed in aspects of variations, invariant features and temperature model. These normalizations improve the robustness of thermal images on the variations of ambient temperature and enhance the recognition performance when the system is under abnormal state. In addition, the improved normalized coefficient is put forward to confirm the effect of the normalization methods from the angle of image similarity.