Research On Occlusion Face Detection Based On Adaboost Algorithm

With the rapid development of artificial intelligence,the requirements for digital image processing are getting higher and higher.As an important field in digital image processing,face detection also has a tendency toward high efficiency and high detection accuracy.At the same time,face detection is also the first step in face image processing,which has a crucial impact on subsequent face recognition and facial expression analysis.Face detection has also become a research hotspot in digital image processing in recent years.This paper studies and improves the Adaboost algorithm,and explores the face detection of occlusion images.The main contents are as follows:By training the training and detection principle of Adaboost algorithm,the training part of the algorithm is improved.Firstly,this paper expounds the principles of Haar-like features and integral images,and discusses methods for training weak classifiers,strong classifiers and the cascading of strong classifiers,and introduces the principle of face detection and combining the detection result windows.Secondly,this paper discusses methods to optimize training samples and reduce the number of training features.Finally,for the problem of too long training time,this paper proposes a training method for double threshold weak classifiers.The experimental results show that the double threshold method has lower accuracy than the traditional method in terms of detection performance,but the recall rate has been significantly improved.In terms of training time,when the number of training sample images is huge,the double threshold method has obvious advantages in efficiency over the conventional method.The double threshold method can effectively reduce the training time under the condition of ensuring the detection performance.This paper uses a new processing method for occluding face images,a POOA(Positioning the optimal occlusion area)face occlusion area positioning algorithm based on saliency detection is proposed.Firstly,the image is processed by the cluster-based saliency detection algorithm,which increases the difference between the face occlusion area and the face area.The POOA algorithm is used to locate and mark the face occlusion area.Then,the repaired face image is obtained by using the Robust PCA algorithm,so that the information of the face occlusion area is approximately obtained.Finally,according to the positioning information and the repair information of the face occlusion area,the positioned face occlusion area is filled to obtain the final face repair image.The experiment of Adaboost algorithm shows that the proposed method can improve the precision and improve the recall higher than the unprocessed method.