Skip to main content

​
Overview

PDQ (Photo-DNA Quality) is a perceptual hash developed by Facebook (now Meta) that produces a 256-bit hash robust to JPEG compression, rescaling, and minor edits while remaining fast enough for large-scale deduplication. The algorithm applies a Discrete Cosine Transform (DCT) to a 64×64 grayscale image, extracts a 16×16 coefficient block, and thresholds it against the median to produce a binary hash.

​
When to Use

Use PDQ when you need:
  • High robustness to JPEG compression and minor edits
  • Large-scale deduplication of images (designed for production use)
  • 256-bit hashes for better discrimination than 64-bit alternatives
  • Industry-standard hash used by content moderation systems
PDQ is particularly effective for detecting duplicate photos, memes, and content variations at scale.

​
Constructor

func NewPDQ(opts ...PDQOption) (PDQ, error)

​
Available Options

  • WithInterpolation(interp Interpolation) - Sets the resize interpolation method
  • WithDistance(fn DistanceFunc) - Overrides the default Hamming distance function

​
Supported Interpolation Methods

  • NearestNeighbor
  • Bilinear (default)
  • Bicubic
  • MitchellNetravali
  • Lanczos2
  • Lanczos3
  • BilinearExact

​
Usage Example

package main

import (
    "fmt"
    "github.com/ajdnik/imghash/v2"
)

func main() {
    // Create PDQ hasher with default settings
    pdq, err := imghash.NewPDQ()
    if err != nil {
        panic(err)
    }

    // Hash an image file
    hash, err := imghash.HashFile(pdq, "photo.jpg")
    if err != nil {
        panic(err)
    }

    fmt.Printf("PDQ hash: %v\n", hash)
}

​
With Custom Options

// Create PDQ hasher with custom interpolation
pdq, err := imghash.NewPDQ(
    imghash.WithInterpolation(imghash.Bicubic),
)
if err != nil {
    panic(err)
}

hash, err := imghash.HashFile(pdq, "photo.jpg")

​
Default Settings

  • Hash size: 256 bits (32 bytes)
  • Resize dimensions: 64×64 pixels (fixed)
  • Interpolation: Bilinear
  • Distance metric: Hamming distance
  • Processing: DCT-based with Jarosz filter

​
Technical Details

The PDQ algorithm:
  1. Resizes the image to 64×64 pixels
  2. Converts to grayscale
  3. Applies Jarosz box filter (window=2, reps=2)
  4. Computes DCT on the filtered image
  5. Extracts top-left 16×16 DCT coefficients
  6. Thresholds coefficients against their median
  7. Produces a 256-bit binary hash

​
Comparison

PDQ hashes are compared using Hamming distance by default. Lower distances indicate more similar images. 
dist, err := pdq.Compare(hash1, hash2)
if err != nil {
    panic(err)
}

if dist < 20 {
    fmt.Println("Images are very similar")
}

​
References

PDQ is optimized for real-world photo matching and is used by major platforms for content moderation and copyright detection.

Build docs developers (and LLMs) love

Get started for freeTalk to us