Feature Point Extraction Of Dog Face Based On ERT Algorithm

In recent years,China's comprehensive national strength has developed rapidly,and the size of the pet market has also expanded.Dogs,as the most popular pets,play an impor-tant role in the pet market,which is why it is particularly important to further study the dog's face.In the further study of the dog's face,dog face recognition is the most important one.Referring to the face recognition system,in order to realize the recognition of the dog face,the conventional method is to extract the feature points of the dog face,and then implement the dog face recognition based on the extracted feature point training model.Although the feature point extraction of the face has been relatively mature,there is little research on the feature point extraction of the dog face.In this paper,with reference to face recognition,the dog face feature point extraction in dog face recognition is studied.It is found that different dog faces are difficult to establish a unified model due to the difference in shape.Therefore,this paper proposes to establish different features for different dog faces.The point extraction model is given,and an innovative feature point extraction method is given to realize the input of an unknown face dog face image,which can extract the feature points that are most suitable for the dog face.In this paper,the HOG feature is combined with the SVM method to realize the dog face detection.Then,for a given variety,the ERT algorithm is used to realize the feature point extraction of the dog face of the breed.Finally,based on the previous two steps,two sets of dog face feature points are designed.The method of extraction,the first one adopts the idea of enumeration method,compares the input picture with the varieties in the target detection model one by one,selects the possible variety type sets,and then extracts the feature points by using the model corresponding to the varieties in the collection.Fi-nally,according to the pixel gray value of the feature point,the Euclidean distance between the average pixel gray value of the corresponding variety is used to determine the optimal feature point of the input picture dog face;the second method adopts the clustering idea and will all The training pictures are clustered according to the HOG features of the face screenshots,and the varieties included in the k subclass are determined,and the detection models are respectively established for the k subclasses.Next,compare the input image with the detection model of all subclasses to determine the subclass in which it is located.After that,the feature points are extracted using the model corresponding to the variety contained in the subclass.Finally,according to the pixel value of the feature point,the Eu-clidean distance between the average pixel value and the corresponding pixel value is used to determine the best feature of the input picture dog face.In order to test the performance of the two extraction methods,this paper considers 13 varieties of Keji,Husky,Samoyed,Shiba Inu,Bomei,Fadou,Demu,Bull Terrier,Bully,Golden Retriever,Alaska,Labrador and Doberman.Each model has 30 maps,and each graph has a test set of 41 feature points for testing.The first method has an MSE of 11.7217 on the test set and a total time of 1652 seconds.The second method has an MSE of 16.6276 on the test set and a total time of 409 seconds.It can be seen from the performance on the test set that method one is more accurate than method two,and method two is shorter than method one.It is concluded that the first method is more suitable for scenes with high pre-cision requirements and low time requirements,such as extracting dog face feature points for dog face recognition;method 2 is more suitable for scenes with high time requirements and low precision requirements.For example,online extraction of dog face feature points for dog face beauty.