PHP class for comparing two images and determine how much they similar, in percentages.

determine-images-similarity.php

<?php
/**
 * Compare two images and determine how much they similar, in percentages.
 */
class ImagesComparator
{
    private const ACCEPTED_IMAGE_TYPES = [IMAGETYPE_JPEG, IMAGETYPE_PNG];

    /**
     * Calculate the images similarity in percentages.
     *
     * Based on calculation of the hammering distance between color bit values of two images.
     *
     * What is about the snapshot size? Images comparison makes a lot of calculations based on pixels color.
     * These calculations will give roughly the same result for the image in a different sizes, but the bigger image
     * required more time for calculation. So it is make sense to use a resized images instead originals, which is
     * called here the snapshots. Recommended value in the range of [8, 100] px.
     *
     * The smaller snapshots size, the rough the comparison you"ll get (in theory).
     *
     * @param string $file1 absolute path + file name to the first image
     * @param string $file2 absolute path + file name to the second image
     * @param int $snapshotSize snapshot with (or height) in pixels. The snapshot is a square image, always.
     * @return int|null percentages of similarity
     */
    public function compare(string $file1, string $file2, int $snapshotSize): ?int
    {
        $image1 = $this->prepareSnapshot($file1, $snapshotSize);
        $image2 = $this->prepareSnapshot($file2, $snapshotSize);

        if (!($image1 && $image2)) {

            return null;
        }

        $colorsMap1 = $this->buildColorsMap($image1);
        $colorsMap2 = $this->buildColorsMap($image2);

        $deviationMap1 = $this->determineDeviationFromMeanValue($colorsMap1);
        $deviationMap2 = $this->determineDeviationFromMeanValue($colorsMap2);

        $similarityPercentage = $this->calcImagesSimilarity($deviationMap1, $deviationMap2);

        imagedestroy($image1);
        imagedestroy($image2);

        return $similarityPercentage;
    }

    /**
     * Create resized image resource from the source file. Apply grayscale filter.
     *
     * Accept jpeg or png source only.
     *
     * @param string $source absolute path + file name to the image
     * @param int $snapshotSize
     * @return resource|null - image resource identifier or NULL if failed
     */
    private function prepareSnapshot(string $source, int $snapshotSize)
    {
        if (!file_exists($source)) {
            return $this->fail("File not found: " . $source);
        }

        $heap = getimagesize($source);
        $imageType = $heap[2];
        $mimeType = $heap['mime'];

        if (!in_array($imageType, static::ACCEPTED_IMAGE_TYPES)) {
            return $this->fail("Unsupported image type. Its MIME type is " . $mimeType);
        }

        $sourceImage = $this->createImage($source, $imageType);

        if (!$source) {
            return $this->fail("Failed to create image resource");
        }

        $snapshot = $this->resizeImage($sourceImage, $snapshotSize);

        if (!$snapshot) {
            return $this->fail("Failed to resize the image");
        }

        $isFiltered = imagefilter($snapshot, IMG_FILTER_GRAYSCALE);

        if (!$isFiltered) {
            return $this->fail("Failed to apply grayscale filter");
        }

        return $snapshot;
    }

    /**
     * Create image resource. Supports jpg and png only.
     *
     * @param string $source absolute path + file name to the image
     * @param string $imageType image type, see php::IMAGETYPE_* constants
     * @return resource|null - image resource identifier or NULL if failed
     */
    private function createImage(string $source, string $imageType)
    {
        switch ($imageType) {
            case IMAGETYPE_JPEG:
                return imagecreatefromjpeg($source) ?: null;
            case IMAGETYPE_PNG:
                return imagecreatefrompng($source) ?: null;
            default:
                return null;
        }
    }

    /**
     * Resize the source image to specified size. The result will be a square image despite of original aspect ratio.
     *
     * @param $sourceImage
     * @param int $resizeTo
     * @return resource|null
     */
    private function resizeImage($sourceImage, int $resizeTo)
    {
        $resizedImage = imagescale($sourceImage, $resizeTo, $resizeTo);
        return $resizedImage ?: null;
    }

    /**
     * Build the map of all pixels color in the image
     *
     * @param resource image resource identifier
     * @return array
     */
    private function buildColorsMap($image): array
    {
        $width = imagesx($image);
        $height = imagesy($image);

        $colorsMap = [];
        for ($w = 0; $w < $width; $w++) {
            for ($h = 0; $h < $height; $h++) {
                $rgbIndex = imagecolorat($image, $w, $h);
                $colorsMap[] = $rgbIndex & 0xFF;
            }
        }

        return $colorsMap;
    }

    /**
     * Determine a deviation of each pixel from the colors mean value.
     *
     * If a color is bigger than the mean value of colors - it is 1, other vise it"s 0.
     *
     * @param array $colorsMap
     * @return array
     */
    private function determineDeviationFromMeanValue(array $colorsMap): array
    {
        $colorMeanValue = array_sum($colorsMap) / count($colorsMap);

        $deviations = [];
        foreach ($colorsMap as $color) {
            $deviations[] = (int)($color >= $colorMeanValue);
        }

        return $deviations;
    }

    /**
     * Calculate the images similarity
     *
     * Calculate the Hamming distance between two maps of color deviations. The relation of that distance to the map
     * count can be interpreted as the images difference in percentages.
     *
     * @param array $deviations1
     * @param array $deviations2
     * @return int percentages of similarity
     */
    private function calcImagesSimilarity(array $deviations1, array $deviations2): int
    {
        $distance = 0;

        // Note: it doesn"t matter which exactly map to take, they all have the same size.
        $mapSize = count($deviations1);

        // Don't use this. It gives a wrong result for 256 elements per array.
        // $distance = count(array_diff($deviations1, $deviations2));

        for ($i = 0; $i < $mapSize; $i++) {
            if ($deviations1[$i] !== $deviations2[$i]) {
                $distance++;
            }
        }

        return 100 - $distance * 100 / $mapSize;
    }

    /**
     * Report about fail and return nothing.
     *
     * In the perspective, you can replace `php::echo()` here with logging or throwing exception.
     *
     * @param string $message
     * @return null
     */
    private function fail(string $message)
    {
        echo $message;
        return null;
    }
}

Usage example:

$comparator = new ImagesComparator;
for ($snapshotWidth = 8; $snapshotWidth <= 100; $snapshotWidth += 8) {
    $similarity = $comparator->compare('/path/to/first_image.jpg', '/path/to/second_image.jpg', $snapshotWidth);
    echo "Similarity: {$similarity}% on snapshot width {$snapshotWidth}px" . PHP_EOL;
}

It is not necessary to use the cycle with many snapshot widths. It's just a demo. On the practice you should choose acceptable width just one time and use it everywhere.

The quote from the class:

What is about the snapshot width? Images comparison makes a lot of calculations based on pixels color. These calculations will give roughly the same result for the image in a different sizes, but the bigger image required more time for calculation. So it is make sense to use a resized images instead originals, which is called the snapshots. Recommended width value in the range of [8, 100] px.

The smaller the size, the rough the comparison you'll get (in theory).

I found a theoretical basis for this solution.

Now I believe my resizing improvements were redundant. It's not bad, but useless. I have no time to refactoring this class one more time. And I don't want to do that, because here is a vendor library that provides the same logic using Jenssegers\ImageHash\Implementation\AverageHash and three more solutions. I suggest to use this library.