Similar image finder

Question

I have several thousand drawings of small parts and blueprints for which I need to be able to iterate through and search. The goal is be able to give the software an example image, then have the software search the thousands of images in a directory for anything similar, and return 20 or so matches.

I'm not looking for something which find exact image matches, but instead can find similarly shaped objects.

Answer 1

AllDup allows you to find similar images.

The documentation can be found over at https://www.alldup.de/alldup_help/search_similar_pictures.php

It is freeware.

Answer 2

Depending upon how enthusiastic you are at writing your own - you can use the image analysis features built into the Wolfram Language.

This is used in the Image Identification Project which aims to improve machine vision and learning, although that project won't help in your use case.

There is a StackExchange site dedicated to the use of Wolfram. They already have questions about using the image analysis features.

I repeat my initial comment - this isn't an off-the-shelf solution but it should be suitable for your needs.

Answer 3

You can use LEADTOOLS Imaging Pro SDK technology in your application. https://www.leadtools.com/sdk/products/imaging-pro

Imaging Pro SDK - $795 (Price for Developer Seat (08/2021))

You can leverage the CorrelationCommand class, which allows you to match an image containing a portion of the source image to the source image according to the measure of correlation.

DISCLOSURE: I am an employee of the company offering this toolkit.

Here is some sample code:

List<string> files = Directory.EnumerateFiles(@"DIRECTORY_CONTAINING_IMAGES_THAT_COULD_MATCH_TO_PART_OF_SOURCE_IMAGE").ToList();

using (RasterCodecs _codecs = new RasterCodecs())
using (RasterImage _srcImage = _codecs.Load(@"FILE_PATH_TO_MAIN_SOURCE_IMAGE"))
{
    foreach (string file in files)
    {
        using (RasterImage _dstImage = _codecs.Load(file))
        {
            CorrelationCommand command = new CorrelationCommand();
            // Set the CorrelationImage equal to the thousands of small image parts you are iterating through
            command.CorrelationImage = _dstImage;
            // The threshold indicates the measure of association required to consider the two images/areas to be correlated
            command.Threshold = 70;
            // Value that represents the step size in the X direction (along image width), in pixels. For best results, use 1
            command.XStep = 1;
            // Value that represents the step size in the Y direction (along image height), in pixels. For best results, use 1
            command.YStep = 1;
            command.Points = new LeadPoint[90];
            // Pass in the example source image
            command.Run(_srcImage);

            // If correlation is found export image to matched images directory
            if (command.NumberOfPoints > 0)
            {
                string _fileName = Path.GetFileNameWithoutExtension(file);
                string _outDir = @"DIRECTORY_TO_SAVE_MATCHED_IMAGES";
                if (!Directory.Exists(_outDir))
                    Directory.CreateDirectory(_outDir);
                string _outFile = Path.Combine(_outDir, $"{_fileName}.tif");

                _codecs.Save(_dstImage, _outFile, RasterImageFormat.Tif, _dstImage.BitsPerPixel);
            }
        }
    }
}

You can also take the correlation LeadPoint and create a LeadRect, which can be used to draw a bounding rectangle on the source image showcasing where the correlation image(s) is located (See the below screenshot).

Also, note that we have a C# .NET tutorial that uses various LEADTOOLS image processing commands to compare two images to determine how similar they are (See in link below). https://www.leadtools.com/help/sdk/v21/tutorials/dotnet-console-compare-image-data-for-similarities.html