Assessing the quality of acquired images to improve ear recognition for children

Abstract

The use of biometrics to secure the identity of children is a continuous research worldwide. In the recent past, it has been realized that one of the promising biometrics is the shape of the ear, especially for children. This is be cause most of their biometrics change as they grow. However, there are shortcomings involved when using ear recognition in children, usually caused by the surrounding environment, and children can be at times uncooperative, such as moving during image acquisition. Consequently, the quality of acquired images might be affected by issues such as partial occlusions, blurriness, sharpness, and illumination. Therefore, in this paper, a method of image quality assessment is proposed. This method detects whether the images are affected by partial occlusions, blurriness, sharpness, or illumination. This method assesses the quality of the image to improve ear recognition for children. In this paper, four different test experiments were performed using the AIM database, IIT DELHI ear database, and ear images collected by Council for Scientific and Industrial Research (CSIR) researchers. The Gabor filter and Scale Invariant Feature Transform (SIFT) feature comparison methods were used to assess the quality of images. The experimental results showed that partial ear occlusions has less than 16 key points, resulting in low identification accuracy. Meanwhile, blurriness and sharpness were measured using the sharpness value of the image. Therefore, if the sharpness value is below 13, it means that the image is blurry. On the other hand, if the sharpness value is greater than 110, the image quality affects the ex tracted features and reduces the identification accuracy. Furthermore, it was discovered that the level of illumination in the image varies, the higher the illumination effect, such as the value above 100 affects the features and reduces the identification rate. The overall experimental evaluations demonstrated that image quality assessment is critical in improving ear recognition accuracy.