ImageTools/hash/src/main/scala/com/sothr/imagetools/hash/type/PHash.scala

package com.sothr.imagetools.hash.`type`

import edu.emory.mathcs.jtransforms.dct.FloatDCT_2D
import grizzled.slf4j.Logging

/**
  * Complex perceptual hash
  * Uses FFT to get structural data out of the image, which can represent the perception of similarity
  *
  * Created by dev on 1/22/14.
  */
object PHash extends PerceptualHasher with Logging {
	def getHash(imageData: Array[Array[Int]]): Long = {
		//convert the imageData into a FloatArray
		val width = imageData.length
		val height = imageData(0).length
		trace(s"Starting with image of ${height}x$width for PHash")

		val imageDataFloat: Array[Array[Float]] = Array.ofDim[Float](height, width)
		for (row <- 0 until height) {
			for (col <- 0 until width) {
				imageDataFloat(row)(col) = imageData(row)(col).toFloat
			}
		}

		//perform transform on the data
		val dct: FloatDCT_2D = new FloatDCT_2D(height, width)
		dct.forward(imageDataFloat, true)

		//extract the DCT data
		val dctDataWidth: Int = width / 4
		val dctDataHeight: Int = height / 4

		//calculate the mean
		var total = 0.0f
		for (row <- 0 until dctDataHeight) {
			for (col <- 0 until dctDataWidth) {
				total += imageDataFloat(row)(col)
			}
		}
		val mean = total / (dctDataHeight * dctDataWidth)
		//debug(s"Calculated mean as $mean from ${total}/${dctDataHeight * dctDataWidth}")

		/*
		For each pixel, if it is at or above the mean, store it as a one, else store it as a zero
		 */
		var hash = 0L
		for (row <- 0 until dctDataHeight by 2) {
			for (col <- 0 until dctDataWidth by 1) {
				hash <<= 1
				val pixel = imageDataFloat(row)(col)
				//
				if (pixel >= mean) hash |= 1 else hash |= 0
			}

			if ((row + 1) < dctDataWidth) {
				val nextRow = row + 1
				//process each column
				for (col <- (dctDataWidth - 1) to 0 by -1) {
					hash <<= 1
					val pixel = imageDataFloat(nextRow)(col)
					if (pixel >= mean) hash |= 1 else hash |= 0
				}
			}
		}
		debug(s"Computed PHash: $hash from ${dctDataWidth * dctDataHeight} pixels")
		hash
	}
}