I have a service that compares two base64 encoded image strings
My initial attempt revealed that there is differences in metadata while the actual image (JPG) in this case is identical (size,resolution,dimensions,etc).
Is there a way to strip away much of the dynamic metadata so that I can just compare the visual aspect of the image?
Currently, I am using the following...
package converter
import (
"bufio"
"encoding/base64"
"log"
"os"
)
func Base64(path string) (string, error) {
imgFile, err := os.Open(path)
if err != nil {
log.Fatalln(err)
}
defer imgFile.Close()
// create a new buffer base on file size
fInfo, _ := imgFile.Stat()
var size int64 = fInfo.Size()
buf := make([]byte, size)
// read file content into buffer
fReader := bufio.NewReader(imgFile)
fReader.Read(buf)
// convert the buffer bytes to base64 string - use buf.Bytes() for new image
imgBase64Str := base64.StdEncoding.EncodeToString(buf)
return imgBase64Str,nil
}
解决方案
Perceptual Hash is a library to calculate a phash; a hash of an image based on visual characteristics. github.com/carlogit/phash is a golang implementation. It has functions to create and compare two hashes to give a 'distance' indicating how dissimilar two images are.
Out of interest I gave it a try, it's simple to use and effective with some test images. For example:
distance: 0
distance: 2
distance: 32
package main
import (
"fmt"
"log"
"os"
"github.com/carlogit/phash"
)
func main() {
if len(os.Args) < 3 {
log.Fatalf("usage: %s \n", os.Args[0])
}
a := hash(os.Args[1])
b := hash(os.Args[2])
distance := phash.GetDistance(a, b)
fmt.Printf("distance: %d\n", distance)
}
//hash returns a phash of the image
func hash(filename string) string {
img, err := os.Open(filename)
if err != nil {
log.Fatal(err)
}
defer img.Close()
ahash, err := phash.GetHash(img)
if err != nil {
log.Fatal(err)
}
return ahash
}