Face Recognizer Using Python & OpenCV

1. Introduction : Hello, guys. It's good to see you back. Once again, I'm back with a new class related computer vision. Today in this class, we are going to create real time phase recognization with the help of Open Civil Library and Python. Hello. My name is Jonatha Shorter. I am Fullstack web developer Python programmer and online instructor. It can detect multiple phases and identify the pass at the same time using our webcam. Let me show you the demonstration what we are going to build today in this project. As you can see, our program successfully recognize my phase. Also it recognized Prime Minister Modi phase. Then it recognized Donald Trump phase. After that, it recognized Justin Trudeau phase, and at last, it recognized Georgia Meloni phase. Our program is very much capable to detect and recognize phases. To create this project, we are going to use media pipe and phase recognization module. From the next tutorial, we are going to start our coding. We are going to start with importing liabilities. 2. Install libraries and read the image file: So finally, we are in my visor Studio code editor. And as you can see in my current working directory, I already create a Python file name recognizer dot Pi and a folder. A folder name known underscore faces. In that folder, we have some images. We have image of Georgia Meloni. We have image of Naren Romoi. We have image of Justin Trudeau and at last, we have image of Donald Tram. These are all well known pass all over the world, and for our recognizer, we trained our model with those phases. Now, let's back to the recognizer dot Pi five. And at first, I'm going to import the important modules. At first, we need to import OS module, Sim type, import OS. Using this module, we can access our working directories, otherwise, the folders. Then we need to import another module called CV two. This is the main module. Simu type, import, cv two. With that, we need to input another module. But first, we need to download this module, and our module name is phase recognization. Let's open the Windows Power Shell and try to download this module. To download this module, just you need to type, peat, install, and our modulem is phase recognization. Then you need to hit Enter. After hit Enter, as you can see, requirement already satisfied because I already download this module, but you might face some error because of this module, because this module come with some dependencies such as Nampi, Pillow and DL. But in your case, you can get error because of this DL module, and we need to install this module manually. Let me show you how we can install this module. To resolve the problem, just open your browser and search delete Well package download link. Then it provides you this link, this Github link, and you need to open this Github link. Delete compiled binary dot HL file and you need to download this Wheel file. If I click on this file, let me show you, as you can see, here it provides multiple HL file for multiple Python version. If you're using 3.10 version, then you can use this Wheel file. If you use 3.7 version, then you can use this Well file. I am using 3.11, so I download this one. To download it to download it, just select your Windows version. I'm going to use this one and I click on it. After that, you need to click on Download Raw file. After click on Download button, as you can see, it download to this folder, this Download folder. It download to this download folder. Now, let me show you how we can install this WL file using Windows Power Shell. At first, you need to copy this file path. I copy the exact file path, and I'm going to open my Windows Power Shell. And as you can see, I on in my Windows Power Shell. To download this WL file, first, you need to type, peep, install, then you need to use space and you need to use dowel codes. Then inside the dowel codes, you need to put the file path. After that, you need to use Ein, then you need to provide the file name. To provide the fine name, I'm going to copy the exact fine name. So from here, I'm going to copy the exact file name with the extension and I'm going to paste it here, Control V, and we need to hit Enter to download this module in our system. Just hit Enter and it's going to compute your download process. I already download it, so I'm not going to download it again. This is how you can download any Hill file in your local machine. Now, let's get to the code editor. Then after install the DL module, now you are able to install phase recognization module in your system. So just type PIP install phase recognization. You need to follow this method only if your system not able to download D module automatically. Let's import the phase recognization module in our five, some type Import and our module name is phase recognization. Apart that, we need to install another module, which is media Pipe. To install media pyte module, you need to type this command. Pep Install OpenCb Python space media Pipe. If you hit Enter, as you can see, requirement already satisfied because I already download this module in my system. Let's input the module in my system. Some type import, and I'm going to import media Pipe as MP. That's it. After import all the necessary modules, we need to load the known faces. I want to say, we need to load all the known faces such as Meloni Modi, Trudo and Tram. Remember, you can use as much pass you want. It's up to you. So here I'm going to type a single command load known phases. After that, I'm going to declare two variable. To store the known phases, we need to define two variable, basically two list. Our first variable name is known unc phase, under sco, encodings, equal to I empty list. Then I'm going to define another variable which is known phase name known under sco, phase, under sco encodings, underscoename. Names equal to blank. Our first variables store the numerical encodings of each known pass and the second variables store the names named corresponding to each pass. In our first list, we are going to store the numerical encodings and in our second variable, we are going to store the name of the images, and it's going to get all the names from the fine name such as Meloni Modi Trudeau tram I' going to take the image name as a person name. Next, we need to look through all files in this known phase folder. For that, we are going to use for loop. Let me show you how. Sum rotyeFFlme in s dot, I'm going to use is DA function. Then inside the round reces I'm going to provide the folder name which is non phase, K Wn underscore phase, nonpass. Basically, this four loop going to look through every file inside the nonpas folder. Then inside the four loop, we need to construct the full file path for every image. I want to say here, file name represent images, the image file, which is inside this nonpas folder. We look through every image. So we need to construct the full file path, so I'm going to define another variable called PAT. But equal to here, I'm going to use Join function. Type Os path dot join. Then inside the round recess, first, we need to provide the folder name, and our folder name is non faces. Non faces and apricoma we need to provide the we need to provide the file name. I'm going to pass the final name variable. File name. It's going to join the non pass and the current file to get the full file path of the image. Suppose for Meloni, it's going to get the folder name non phase, then it's going to add the file path, meloit JPG. Next, we need to load the image in phase recognization module. For that, I'm going to define another variable called image called image. Image equal to here, I'm going to type phase recognization dot and from phase recognization, I'm going to call a method called load image file. Load image file. Then inside the roundres I'm going to provide the path variable. Basically, it load the image file into memory in a format that pace recognization library can process. Basically, behind the scenes, it used Nampi alley. Next, we need to get the phase encodings. For that, I'm going to define another variable called encodings. Encodings equal to phase recognization. Dot, I'm going to use from phrase reorganization, I'm going to use phase encodings. Then inside the roundresses, we need to provide the image variable image. And I'm going to set this five. It is try to find all phases in the image and generate a 128 decimal encoding for each images. Basically, it return a list. Even if there is only one phase, still it return a list. Now if the pace is found, we need to set the phase encoding to this list and the pas name to this list. For that, here, I'm going to use IP condition. If encodings, then inside this IP condition, then I'm going to append the encodings to this list. So types, non phase encodings dot append function, append then inside the round resis, I'm type encodings. Inside the square recess, I'm going to provide zero. Basically, it takes the first phase encoding found in this image and append it to this list, non phase encodings. Next, we need to extract the image finally. For that, we are going to use split text method. Here I'm going to type Os, dot, butt, split text. Then inside the round braces, I'm going to provide the file name. Then inside the square braces, I'm going to pass zero. Basically, it's going to split the file into tupel such as melon dot JPG. Meloni is one tupel item and dot JPG is another tapel item. Here we pass zero, I mean we are going to extract the first one, so it's going to extract only the name, meloni not the extension. That's why we use zero hair because we need to extract the name only. And then we need to set the names in this list. For that, we need to use append method again. Summary type known phase names dot append method. Then inside the roundreses we need to append the names, but before I'm going to store, I'm going to store it in a variable, which is name sum type name here, name I'm going to provide this variable name inside this append method. Name, and I'm going to set this pile. Basically, this code scan through the known phase directory. And for each image, it's going to load it, extract the first phase encoding, and set the encoding, also the file name to the list items to this list. And make sure all the images you use, they have a clear phase. If you upload an image where you have multiple phases on that image, then it's going to extract only the first phase. It not going to recognize all the phase from the image. That's why I will recommend to upload those type of images which have only single phase, like the ones. This is the first part of this tutorial. And the next part of this tutorial, we are going to working on media Pi phase detection setup and we are going to start the camera. Thanks for watching this video, Stay tuned for the next part. 3. MediaPipe and face detection setup: Hello, guys. Good to see you back. This is the second part of this tutorial. And here, I'm going to say media pipe phase detection. So at first, I'm going to type the single command, which is media pipe, phase detection setup. After that, here, I'm going to define two variable, and our first variable name is MP, media pipe, phase, underscore detection. Equal to aminotype P dot solution phase under sco detection, phase Underscoe detection. After that, we need to create the drawing utiles. For that, I'm going to define another variable which is MP underscoe drawing, equal to p dot solution, drawing utiles, underscoe utiles. Now the is, what is the meaning of this two variable? What can you do with that? In our first variable MP phase understood detection, we load the media Pi phase detection solution, and in our second variable, here we provide utility function to draw phase landmark on images. Now, let's start the camera. For that, I'm going to define another variable, and our variable name is cap, CAP, Cap otherwise captured equal to here we need to call CV two dot video capture method, video capture and inside the round brass, we need to pass the source. As a source, I'm going to pass zero because I have only one webcam in my system. After that, we need to initialize the phase detection module with a minimum confidence. For that, I'm going to type with MP, underscore phase detection dot from here, I want to run phase detection. Phase detection. Then inside the round roses here, I'm going to set minimum detection confidence. Man underscoe detection, underscoe confidence, equal to, and I'm going to set it 0.6. Aber that I'm going to type as Phase detection, phase underscore detection. Then inside it, here, I'm going to run a while loop. Here I will type y, cap dot is opened, is opened. Then inside this while loop, the ins this while loop here, I'm going to declare two variable. First one is for Bullen variable and second one is for the image frame, and you know the process. I'm type RET, comma, and to stort the frame mut frame, equal to cat read method, Cap dot read. Then I'm going to say p condition. If it is not true, I want to say I I not Bian value is true, then inside this IP condition, I want to break the loop. Pick, that's it. Now the question is what I basically did here. At first, we open our default web camera and save it in a cab variable. After that, He we initialize phase detection model with minimum confidence of 0.6 and her use with statement because it is ensure proper resource management. I automatically close the model when it done. Then I read the camera frames continuously using Hilou. Cab dot is open method, checks if the video stream is active, and this is for Boolean value, the frame variable capture image from the webcam and then see a condition. If no frame is received, then it going to exit the loop. Somehow our camera connection is break, then it going to break the loop if there is no footage. After that, inside this while loop, I'm going to set a IP condition. So type I CV to do wait key. Inside the round process, I'm going to wait for ten millisecond and and zero xFF equal to equal to ORD method, RD. Inside the roundresses, I'm going to pass Q, then I want to break the loop. After that, outside the while loop, I'm going to release the catcher. Cap release method, cab dot, release. Also, I'm going to destroy all the window after release the programs type CV to dot, destroy all Windows. Now. Now inside this while loop, I'm going to convert BGR to RGV frame because by default, OpenCV read our frame in BGR mode. But for phase detection, we need RGV frame. So here are our type RGB underscore frame. RV frame equal to CV two dot, CVT color. Then inside the roundresses, first, I'm going to provide the actual frame as a source, some type frame, then a coma, here I'm going to pass CV two. We need to pass the color code, which is color BGR RGV because we are going to convert big to RGV frame. That's it. Next, we need to extract the phase location, also the phase encodings. For that, I'm going to define two more variable. In our first variable Imo type phase underscoe locations. Phase locations equival to phase detection, dot, dot, phase location method, phase underscoe locations. Then inside the round verses here I will pass the RGV frame GV frame. After extract the phase location from the RGV frame, also, we need to extract phase encodings. For that, I'm going to define another variable phis underscoe encodings. Fit encodings equal to herm type phase recognization, from phase recognization, I 12, sorry, also you need to run phase recognization, not detection, phase recognization. This one. Using phase recognization module, here we extract the phase location. Then using phase recognization module, we are going to extract the phase encodings. Here, I'm going to tie phase recognization, dot, I'm going to type phase encodings. This method. Then inside the round resses here, we need to pass first, we need to pass the RGV frame, type RGV frame, comma, then I'm going to pass this variable phase location, phase locations, this one. I'm going to set this one. Now let me explain to you basically what basically it going to do. Phase locations retrat the coordinate of all detected phases in the frame, and our phase encoding computes 12 dimension phase encodings for each detected phase. Now we have phase location and phase encoding. Next, we need to recognize the phase. For that, we are going to use a follow, but not in this tutorial. And the next part of this tutorial, we are going to covert the remaining part. Thanks for watching this video stay tuned for our next. 4. Main loop that going to recognise the face: Hello guys. Good to see you back. This is the third part of this tutorial, and here we are going to run a loop. We are going to run the main loop that going to recognize the phase. For that, we are going to use follow, but before, I'm going to type a single and comment, and I'm going to recognize phases. After that, we need to start the follow. We need to run a four loop through each detected phase. So I'm going to type for inside the round presses, I'm going to take some variable for the coordinates, some type top, comma right. Bottom, t. After that, phis encoding. Phase encodings. Fizz encoding in Z Z IP Z inside the roundress hemmed pass, phase, location, face locations, fizz encodings, phase encodings. Then inside this four loop, but let me explain this fo loop what I just did here. What is the meaning of this four loop? Basically, here we look through all detected phase. These loops iterate over each detected phase. Here, phase locations contain bounding box coordinate of phases, which found in the frame, and phase encodings contain facial feature vectors for each phases. Basically phase recognization module use 128 vector points to identify phase, and then we use Zip method. It going to pairs each phase locations with its encoding, and top right, bottom lapped are the coordinates used to draw a rectangle around the face. After that, inside this four loop, we need to find the matches to compare the faces. For that, I'm going to declare another variable called matches. Match is equal to here I'm going to use a method called phase compare from phase recognization. So ti phase recognization, dot compare phase, compared phases. Then inside the round ss, here, I'm going to pass total two variable. Non phase encodings, non phase encodings and phase encoding. Phase encoding. He compare the current phase encodings known phase with non phase encodings. As you know, in our non phase encodings, here is told all the phase coordinates and their features from these known phase images such as Meloni, Modi, Trudeau, Tramp, et cetera. It's going to compare the current phase with the known phase images. If you do not found any match, then I want to name it unknown. Sutname equal to unknown inside the double course not unknown. It will be overwritten if the known phase is not found. Basically, it is the default label name. And next, we need to use the shortest distance for the best match. For that, we need to calculate the distance. So here, I'm going to type another single and command and I'm going to pass use the shortest distance for the best match. I'm going to type a variable name and our variable name is phase distances. Phase undersco distances to calculate the distance, we need to use a method from phase recognization, which is phase distance, sum type phase recognization, dot phase distance method, phase distance. Then inside the round recess here we need to compare with non phase encodings with current phase encodings type, non phase encodings. Comma, phase encodings, phase encodings. Phase encoding. Basically, this function calculates Ecledean distance between the current phase and the all known phase. Similar distance mean more similarity, and we use this method to find the best match from the known phases. Now we need to run an I condition to ensure there is at least one known phase to compare with. Here I'm going to type I lane inside the round versus phase distance, phase distance, greater than zero, then insert this IP condition, then inside this IP condition, we need to get the index of smallest distance for the best possible match. So I'm going to define another variable, and our variable is based undersce match, underscore, index equal to here I'm going to run a method. I'm going to tie phase distances dot Arg mean, this one. This method going to return the index of smallest distance for the best possible match, and we set it in this variable. Now, if the match is found from the best match index, then we need to replace the unknown name with this file name. I want to say this filename. If it is tram then we need to name it Tram if it is Trudeau then we need to name it Trudeo. So for that, I'm going to use HIP condition inside this IP condition. Type I type IP matches inside this square basis, I'm going to pass best match index, then inside this HIP condition, I'm going to overwrite the name variable. Then the name become equal to known phase names inside the square bases from it, I want to put the best match index, best match index. If it found the match, then it's going to put the name. Then we need to draw the rectangles. We need to draw the rectangles. Around the face. For that, outside this IP condition. Outside this IP condition, I'm going to use rectangle method, similar tit CV two dot rectangle. And inside the roundresF I'm going to provide the frame name in which frame we are going to draw draw the rectangles, which is frame. After that, we need to define the position. As you know, we already extract lap top and bottom position to draw the rectangle. At first, we need to provide the top left corner. Here inside the round braces, I'm going to provide the lab value lab and then I'm going to pass to value, top. Next, we need to define the other corner, which is right bottom, which is right bottom. Inside the round braces, I'm going to type right bottom. After that, after that, we need to define the color of this rectangle. Inside the round braces, I'm going to use this colored code. B, for blue, I'm going to use zero for G for green, I'm going to use 255 and for red, I'm going to use zero again. Then we need to provide the thickness of this border, which is two. After that, draw the rectangle, also, you need to put the text, the name of this person. I'm going to use put text method, some type CV two dot, put text. Inside the roundssF we need to provide the name frame in which frame we are going to put the text. Then we need to provide the text, which is name. After that, we need to define the position to define the position of the text. Here inside the round brrss we are going to start the text from the left side, lift and comma. This is for Xxs and for the xs, from the top POP top, I'm going to minus ten film. Autocoma, we need to define the font face. And here, I'm going to use CV two dot, CV two dot, font hoch simplex, this font. Then we need to define the font scale. Here I'm going to use 0.8. After that, we need to define the color. We need to define which color we want to show the font, and I'm going to use the same color. So we use zero comma 255 comma zero. Cream colored and then they want to provide the thickness of the fat which is two, is good for each, and that's it. After that, we need to display. We need to display the frame using IMF method. Outside the fall loop here, I'm going to display the face. So tie CV two taught IMF Mathod. I am sure that inside the round dresses at first want to provide a name for the window and its name is face recognization. And then Apacmo we need to display the frame, the actual frame which we want to display. And I'm going to sub this file. So we are done. And before I run this code in my terminal, I also want to put my own face in this non phase list in this folder. So I put my photo on this list. This photo is taken from my webcam, and I'm going to use this photo also. Now we have total five phase to identify. So let's battle the code and set this file and run this code in our terminal. Now, let's run the code. So I'm going to hit the Run button. After I hit the run button, we need to wait for some time to start the webcam. Afterwards sometime, as you can see, it start my webcam. As you can see, instantly it recognize my face. It creates a square box around my face, and above the square box, it print my name, joint. And if I try to show you some other faces, so in my keyboard, I already print the faces. And if I show it in front of camera, as you can see, it's also recognized other faces. It recognized the other world leader faces such as Modi. And if I ship to the next phase, so I'm going to change the face, as you can see, also it recognized President Trump face. At the same way, if I change the image, then you can see it recognized Canada ex Prime Minister Justin Trudeau. As you can see, we track, also it recognized the face, and then come our last pace Georgia Malone. Also it recognized her special features, so we are nicely done our job. I hope now it's clear for you, how we can build our own face recognization model using Open CP. Thanks for watching this video. Stay tuned for our next project.