Principle analysis of face recognition technology based on face recognition lock

人臉識別鎖人臉識別技術原理分析

Face recognition is mainly divided into three processes: face detection, feature extraction and face recognition.

Face detection: face detection is to detect and extract face image from input image. Generally, Haar feature and AdaBoost algorithm are used to train cascade classifier to classify each image. If a rectangular region passes through a cascade classifier, it is identified as a face image.

特征提?。禾卣魈崛∈侵竿ㄟ^一些數字來表征人臉信息，這些數字就是我們要提取的特征。常見的人臉特征分為兩類，一類是幾何特征，另一類是表征特征。幾何特征是指眼睛、鼻子和嘴等面部特征之間的幾何關系，如距離、面積和角度等。由于算法利用了一些直觀的特征，計算量小。不過，由于其所需的特征點不能精確選擇，限制了它的應用范圍。另外，當光照變化、人臉有外物遮擋、面部表情變化時，特征變化較大。所以說，這類算法只適合于人臉圖像的粗略識別，無法在實際中應用。

Feature extraction: feature extraction refers to the representation of face information by some numbers, which are the features we want to extract. The common face features are divided into two categories, one is geometric features, the other is characterization features. Geometric features refer to the geometric relationships among facial features such as eyes, nose and mouth, such as distance, area and angle. Because the algorithm uses some intuitive features, the amount of calculation is small. However, its application is limited because the required feature points can not be selected accurately. In addition, when the illumination changes, the face is covered by foreign objects, and the facial expression changes, the features change greatly. Therefore, this kind of algorithm is only suitable for rough recognition of face image, and can not be applied in practice.

　　特征提取：特征提取是指通過一些數字來表征人臉信息，這些數字就是我們要提取的特征。常見的人臉特征分為兩類，一類是幾何特征，另一類是表征特征。幾何特征是指眼睛、鼻子和嘴等面部特征之間的幾何關系，如距離、面積和角度等。由于算法利用了一些直觀的特征，計算量小。不過，由于其所需的特征點不能精確選擇，限制了它的應用范圍。另外，當光照變化、人臉有外物遮擋、面部表情變化時，特征變化較大。所以說，這類算法只適合于人臉圖像的粗略識別，無法在實際中應用。

Representation features use the gray information of face image to extract global or local features through some algorithms. LBP is one of the most commonly used feature extraction algorithms. LBP method first divides the image into several regions, and uses the center value as the thresholding value in the neighborhood of 640x960 pixels in each region. The result is regarded as a binary number.

　　表征特征利用人臉圖像的灰度信息，通過一些算法提取全局或局部特征。其中比較常用的特征提取算法是LBP算法。LBP方法首先將圖像分成若干區域，在每個區域的像素640x960鄰域中用中心值作閾值化，將結果看成是二進制數。

Face recognition: the face recognition mentioned here is a narrow sense of face recognition, that is, the extracted features of the face to be recognized are compared with the features of the face in the database, and the classification is determined according to the similarity. And face recognition can be divided into two categories: one is confirmation, which is the process of comparing the face image with the person's image stored in the database to answer your question or not; The other is recognition, which is the process of matching a face image with all the images stored in the database to answer the question of who you are. Obviously, face recognition is more difficult than face recognition because it needs massive data matching. The commonly used classifiers are nearest neighbor classifier, support vector machine and so on.

　　人臉識別：這里提到的人臉識別是狹義的人臉識別，即將待識別人臉所提取的特征與數據庫中人臉的特征進行對比，根據相似度判別分類。而人臉識別又可以分為兩個大類：一類是確認，這是人臉圖像與數據庫中已存的該人圖像比對的過程，回答你是不是你的問題;另一類是辨認，這是人臉圖像與數據庫中已存的所有圖像匹配的過程，回答你是誰的問題。顯然，人臉辨認要比人臉確認困難，因為辨認需要進行海量數據的匹配。常用的分類器有最近鄰分類器、支持向量機等。

Similar to fingerprint application, intelligent door lock is a mature technology of face recognition. Because in the face recognition intelligent door lock system, the user is active cooperation, can obtain the qualified face in a specific environment. This provides a good input source for face recognition, and can often get satisfactory results.

　　與指紋應用方式類似，人臉識別技術目前比較成熟的也是智能門鎖。因為在人臉識別智能門鎖系統中，用戶是主動配合的，可以在特定的環境下獲取符合要求的人臉。這就為人臉識別提供了良好的輸入源，往往可以得到滿意的結果。