Image Recognition And Classification Based On Machine Learning And The Application Of Gaze Tracking In Single Camera Module

Image processing is any form of processing for photo or video.Images and videos are inputs in this system.The output can be an image,can be a set of characteristics,also can be parameters related to the image.Firstly,import images or videos with an optical device such as a camera or camcorder.Then we analyze the image,summarize data in some way,so that we can find the rules that cannot be seen by the human eyes.For example,the chapter 2 of this article uses this method to find the regular characteristics of the human hand and the images of different cells.Then,the model generated by the regular features is used to recognize the hand and distinguish normal cells from abnormal cells.Besides,in chapter 3,we uses this method to process face images to find face characteristics,and then we can monitor faces.In this paper,we introduce the application of machine learning in the domain of image processing and classification,build a system of gaze tracking and introduce the CLM algorithm.The main research and innovation points in this paper can be summarized as follows:1)We study the framework of Convolutional Architecture for Fast Feature Embedding.We calibrate the photos of normal lymph node and hyperplastic lymphadenitis tissue,as well as normal cervical cells,cervical squamous cell,carcinoma cells and keratinized keratinocytes cells.We modify the code and parameters,train the database and get a model.We can use this model to identify normal cells and abnormal cells.2)We study the framework of Darknet YOLO.We calibrate our own database,modify the code and parameters to train the database by using the framework of Darknet YOLO.Using this model can achieve hand recognition.3)Different from other methods at need external devices or use special light source,we only need a laptop or a desktop computer with one camera.We study the constrained local model algorithm to achieve gaze tracking in single camera module.Using the coordinates of the characteristic points of eyes and the orientation coordinates of head,we can build the model.Then,we get the parameters by calculating linear equations and taking parameters into the equation.Finally,we get the coordinates of gaze points.We can track gaze after calibration.By comparing the coordinates of the fixation point and simulating the mouse,the feasibility of this method is verified.