5-2 15 Naga Venkata Satya Pranoop Mutha

5-2 23 Geovanni West

---

Link to histogram source: https://github.com/GeoSnipes/Big-Data/tree/master/ICP/icp5/documentation/histogram

object SimpleServer {

  def main(args: Array[String]) {

    implicit val system = ActorSystem("my-system")
    implicit val materializer = ActorMaterializer()
    implicit val executionContext = system.dispatcher

    val route:Route = cors() {
      path("get_custom") {
        post {
          entity(as[akka.http.scaladsl.model.FormData]) { formData:akka.http.scaladsl.model.FormData =>

            val imageByte = (new BASE64Decoder()).decodeBuffer(formData.fields.toString());
            val bytes = new ByteArrayInputStream(imageByte)
            val image = ImageIO.read(bytes)
            ImageIO.write(image, "png", new File("image.png"))
            val rep = IPApp.testImage("image.png")
            complete(rep)
          }
        }
      }
    }


    val bindingFuture = Http().bindAndHandle(route, "127.0.0.1", 8080)

    println(s"Server online at http://127.0.0.1:8080/\nPress RETURN to stop...")
    StdIn.readLine() // Run until user presses return
    bindingFuture
      .flatMap(_.unbind()) // trigger unbinding from the port
      .onComplete(_ => system.terminate()) // and shutdown when done
  }
}

The prediction web server is binded to port 8080. Once it receives an image, it calls IPapp which actually runs the various methods to come up with a prediction.

object IPApp {
  val IMAGE_CATEGORIES = List("apple_pie","baby_back_ribs","baklava")

  /**
    * @note Test method for classification on Spark
    * @param sc : Spark Context
    * @return
    */
  def testImageClassification(sc: SparkContext, path: String): String ={

    val model = KMeansModel.load(sc, IPSettings.KMEANS_PATH)
    val vocabulary = ImageUtils.vectorsToMat(model.clusterCenters)
    val desc = ImageUtils.bowDescriptors(path, vocabulary)
    val histogram = ImageUtils.matToVector(desc)

    println("-- Histogram size : " + histogram.size)
    println(histogram.toArray.mkString(" "))

    val rfModel = RandomForestModel.load(sc, IPSettings.RANDOM_FOREST_PATH)
    val p = rfModel.predict(histogram)
    (s"Test image predicted as : " + IMAGE_CATEGORIES(p.toInt))
  }


  def testImage(string: String):String = {
    val conf = new SparkConf()
      .setAppName(s"IPApp")
      .setMaster("local[*]")
      .set("spark.executor.memory", "6g")
      .set("spark.driver.memory", "6g")

    val sparkConf = new SparkConf().setAppName("ImgClassifier").setMaster("local[*]")
    val sc= SparkContext.getOrCreate(sparkConf)
    val res = testImageClassification(sc, string)

    printf(res);
    res
  }
}

IPapp then takes the image and genereates data based on that image in the form of a histogram. It then outputs its prediction using the model generated from Random Forrest.

A few of the predictions given:

• 
• 
• 
• 
• 

Below is the best confusion matrix generated. It has been formatted to make it more readable but it was taken straight from the console. Our model had a prediction accuracy of 64%.

Source Code for Part 1: https://github.com/GeoSnipes/Big-Data/tree/master/ICP/icp5/src/image_classification/src/main/scala