Identify similar images in a folder based on an image from the scanner

Question

I have a question that I can't figure out,I do not know where to begin.I use the following technologies: React Native (mobile application) and the backend application is built with Node Js.

What I'm trying to do is, the user takes a picture of a product, with the help of this mobile application the caption is saved in a directory on the server, let's say [image_phone],I managed to do this part,on the server I also have a directory [app_images] where I have images of products with a specific name.

What I want to do is create a function that will return from the folder [app_images] all the images similar to the one uploaded by the user in the folder [image_phone].

What did I try to do about it:

1. Following a tutorial on the Internet, I tried to solve this problem using the following modules.[jimp, pixelmatch, pngjs]

Code:

const Jimp = require('jimp');
const PNG = require('pngjs').PNG;
const pixelmatch = require('pixelmatch');

exports.compareImage = catchAsync(async(req,res,next)=>{




    const originalFile = imageRoot+'/imagine3.png';
    const compareFile = compareRoot+'/Screenshot_72.png';


    const createBufferImage = async (url) => {
        return new Promise(async (resolve, reject) => {
            await Jimp.read(url, async (err, image) => {
                if (err) {
                    console.log(`eroare la citirea imagini Jimp: ${err}`);
                    reject(err);
                }
                image.resize(400, 400);
                return image.getBuffer(jimp.MIME_PNG, (err, buffer) => {
                    if (err) {
                        console.log(`eroare convertire url in buffer: ${err}`);
                        reject(err);
                    }
                    resolve(buffer);
                });
            });
        });
    };


    const compareImageApp = async (
        capture,
        app_image
    ) => {
        try {
            console.log('> Start compare');
            const img1Buffer = await createBufferImage(capture);
            const img2Buffer = await createBufferImage(app_image);
            const img1 = PNG.sync.read(img1Buffer);
            const img2 = PNG.sync.read(img2Buffer);
            const { width, height } = img1;
            const diff = new PNG({ width, height });
    
            const difference = pixelmatch(
                img1.data,
                img2.data,
                diff.data,
                width,
                height,
                {
                    threshold: 0,
                }
            );
    
            const compatibility = 100 - (difference * 100) / (width * height);
            console.log(`${difference} diferenta pixel`);
            console.log(`Compatibilitate: ${compatibility}%`);
            console.log('< Misiune completa');
            return compatibility;
        } catch (error) {
            console.log(`Eroare la compararea imaginilor: ${error}`);
            throw error;
        }
    };

    compareImageApp(originalFile,
        compareFile
)


res.status(200).json({
    status:'succes'
})
 })

Result:

But the accuracy does not really exist, there are images that do not resemble each other at all, but it gives me an accuracy of over 50%.

2)I tried to solve this problem using another module https://www.npmjs.com/package/rembrandt but the same accuracy problem does not exist.

Now I'm trying to solve this problem using tensorFlow js, to make a kind of image recognition app, something like facial recognition.

Do you have any ideas on how to solve this problem, if you have faced it before.