{"id":11510,"date":"2020-06-27T11:48:08","date_gmt":"2020-06-27T09:48:08","guid":{"rendered":"http:\/\/flaven.fr\/?p=11510"},"modified":"2021-01-24T06:27:21","modified_gmt":"2021-01-24T05:27:21","slug":"improve-a-cms-photos-library-qualification-with-ai-facial-recognition-in-python-to-provide-better-images-search-results-to-users","status":"publish","type":"post","link":"https:\/\/flaven.fr\/2020\/06\/improve-a-cms-photos-library-qualification-with-ai-facial-recognition-in-python-to-provide-better-images-search-results-to-users\/","title":{"rendered":"Improve a CMS’s photos library qualification with AI, facial recognition in python, to provide better images search results to users"},"content":{"rendered":"

My last post was a modest attempt in discovering Python potential with the help of Anaconda. Meanwhile, Artificial Intelligence (AI) was bewitched me \ud83d\ude42 So, I gained confidence using Python in order to explore AI librairies such as NTLK, OpenCV or face_recognition from ageitgey (Adam Geitgey).<\/p>\n

You can find the files in my github account More on https:\/\/github.com\/bflaven\/BlogArticlesExamples\/ in python_playing_with_facial_recognition<\/a><\/p>\n

This post is about a real issue that can be solved by IA. I believe that for any technical topic, rather than exploring AI documentation which in itself could be quite wastful. I always strive to tackle a real use case and apply as much as possible what I want to discover.<\/b><\/p>\n

So, applying this principle, I have spotted facial recognition for pictures. Like many, I am iOS mobile Phone user and I saw facial recognition at work. I start to wonder: “How can apply those facial recognition principles in a backoffice?”. As I am currently a PO for a Backoffice, I am supposed to have a clear “vision” of my product but more than that I know perfectly its weaknesses, especially dealing with pictures.<\/p>\n

The benefit of this simple question allows me not to stress about the details right away and instead quickly jot something down as a placeholder for a conversation between me, myself and I, a soliloquy. The intent was to not forget to ask all the right questions either upfront or somewhere down the road but do not prevent from starting a P.O.C.<\/p>\n

Here is my shortlist that highlights some of the functional and nonfunctional concerns or requirements that I wanted to address. This list has driven all my decisions to obtain a pragmatic result without missing the goal.<\/p>\n

Goal:<\/b> How can apply those facial recognition principles in a Backoffice in order to improve image search results at the end?<\/p>\n

Short actions todolist:<\/b><\/p>\n

    \n
  1. Browse unknown people images directory with Facial Recognition Script<\/li>\n
  2. Detect known people faces among these images set them aside<\/li>\n
  3. Insert these known people images in a MySQL database<\/li>\n<\/ol>\n

    Requirement:<\/b> This Spike will be all made in Python but I do not want to end up with an overzealous shopping list of requirements.<\/p>\n

    Let me explain the situation. In the Backoffice, which I am dealing with, is hosting thousand of “unqualified” images that contains faces of: Donald Trump, Xi Jinping, Angela Merkel, Boris Johnson, Emmanuel Macron, Vladimir Poutine, Recep Tayyip Erdo\u011fan or less kwnown poeple in an european centric point of view: Macky Sall, Rodrigo Duterte, Ramzan Kadyrov, Hun Sen, Narendra Modi, Hassan Rohani, Stevo Pendarovski, Nicol\u00e1s Maduro, Edgar Lungu…<\/p>\n

    Remember that we still need a human intelligence to say who is who? For your information, Stevo Pendarovski, \u0421\u0442\u0435\u0432\u043e \u041f\u0435\u043d\u0434\u0430\u0440\u043e\u0432\u0441\u043a\u0438, is president of North Macedonia, holding the office since 12 May 2019 and he looks a bit like me \ud83d\ude42 or that is just the glasses.<\/b><\/p>\n

    The face of Stevo Pendarovski, \u0421\u0442\u0435\u0432\u043e \u041f\u0435\u043d\u0434\u0430\u0440\u043e\u0432\u0441\u043a\u0438, president of North Macedonia<\/b>
    \n\"Improve<\/p>\n

    The idea is to increase the relevance of research and reduce the sample of images that will be retrieved with a traditionnal textual search based on their name. It will save time, money and resources but also improve user experience. So, user do not get no results at all or improper results in his\/her image search.<\/b><\/p>\n

    The fact, no one wants to qualify each image by opening them one after the other, adding correct caption to improve indexation and by consequence future’s search results. We are talking about more than 1 500 000 pictures. Indeed, the wise choice is to leave it to a computer. <\/p>\n

    This is where the “US cavalry” comes to the rescue in the person of facial recognition library. This library will give you an easy way to detect known faces e.g. list of current state leaders, across all these images.<\/p>\n

    The machine learning is here for that and there are tons of pre-trained models and ready to go libraries that you can use. With this magic wand, you properly scale up your skills and impress your friends or your mother. “Look mummy, I am a data scientist!”.<\/p>\n

    As the facial recognition library will truly analyzed the images, it will prevent you to mix up images search with common namesake or word-sense disambiguation. For instance, you can make a difference between Michelle and Barack Obama or if you type Bush and you are looking for the 2 former US presidents, you won’t get australian bush’s pictures.<\/p>\n

    At the end, as a collateral “damage”, when the requalification will be impleteted in the media database, you’ll be able to improve the UX by adding a filter named such as “Portrait” in source selection dropdown menu. Menu that already contains filters like Editorial, Broadcast, Press agency and let the user select source for his\/her images search.<\/p>\n

    One last thing, on an historical point of view, what conclusion can we draw from this allegedly harmless experience?<\/b><\/p>\n

    In the last century, robotization has replaced blue-collar workers in an automobile assembly line for instance. Let’s say ending mass employment in the first industrial revolution industries. What about our nowadays industries? I guess that IA will certainly replace human intelligence e.g. white-collar workers for tedious and repetitive administrative works such as writing meeting minutes, sorting pictures, sorting documents, answering to questions…. and maybe even more creative works! <\/p>\n

    As I speak to you, it is no more a dystopia! IA is creeping everywhere. Exception made in some companies where people are still working like in caveman age. In this case, the purpose here is to “buy social peace” by keeping obsolete jobs alive! Well, sooner or later, there will be an arbitration between some of your job’s tasks and their possible carry out by Artificial Intelligence (AI). Is there a name for this phenomenon? Anyway, you’ll be facing soon this “intelligentization” or “artificialization” of your job so you’d better be ready! I do not even mention the IA\u2019s usage especially facial recognition in surveillance and security that is already a standard!<\/p>\n

    Never forget that modern capitalism loves creative destruction and see AI as a disrupting factor in many industries. This is such a growing question that there’s even a website called, “Will Robots Take My Job?”. The website’s name speaks for itself. You can look up a job title and see the its likelihood of AI-driven doom. https:\/\/willrobotstakemyjob.com\/<\/a><\/b><\/p>\n

    But, let’s come back to our user story, playing with facial recognition in a very inoffensive way !<\/p>\n

    First, I need to have Anaconda install, see previous post for the installation instructions: https:\/\/flaven.fr\/2020\/05\/python-anaconda-worpress-json-json-schema-messy-post-with-few-practices-and-feedback-from-my-p-o-experience\/<\/a><\/p>\n

    Here is a quick overview on the environment that I am using.<\/b><\/p>\n

    if you do not want to read anymore, I made 2 videos to illustrate this post.<\/b><\/p>\n