/ Images

A RxJava based library using native code to convert images to Lowpoly

A RxJava based library using native code to convert images to Lowpoly

RxLowpoly

An Android library to bring your ordinary photos to life with an awesome crystallized effect.

Introduction

RxLowpoly serves as an improvement over XLowPoly by

  • fixing out of memory crashes by scaling down the image in a
    loss-less manner before processing.
  • providing better quality results by using 4000 as the point count
    by default which provides a good trade-off between speed and time.
  • the higher point count leads to a longer execution period, but it is
    significantly reduced by scaling down the image before processing.
  • provides wider choice of input sources like Bitmap, File, Uri
    or Drawable resource.
  • natively using RxJava for background processing thereby reducing
    boilerplate code on the developer's end.

Lowpoly Samples

Original Image Lowpoly Image
Original Lowpoly
Original Lowpoly
Original Lowpoly
Original Lowpoly

Installation

Download

Add the dependency in your app module's build.gradle file

dependencies {
	...
        implementation "com.zebrostudio.rxlowpoly:rxlowpoly:{latest_version}"
}

That's it!

Library Details

  • RxLowpoly uses JNI with 64 bit support to meet google
    specified requirement for all apps to be 64 bit enabled by August
    2019.
  • Use of JNI enables much faster execution than other similar libraries.
  • Use of Sobel Operator for edge detection.
  • Use of Delaunay Triangulation on the
    result from the sobel operator to construct the final crystallized
    lowpoly effect on the image.

JNI

RxLowpoly uses the Java Native Interface to run native code written in C which provides much faster processing for edge detection using the Sobel Operator and then implementing the Delaunay Triangulation algorithm.

Sobel Operator

The Sobel Edge Detector is a gradient based edge detection algorithm which provides us with separate planes on which the Delaunay Triangulation can be applied.

Delaunay Triangulation

We take a set P of discrete points on an image plane and apply Delaunay
Triangulation
DT(P) to produce triangles connecting 3 points at
a time such that no point in **P **is inside the circum-circle of any
triangle in DT(P). These separate triangles taken together in-turn
provide us with the image having a crystallized effect.

Which leads to the resultant crystallized image as :-

Usage Examples

Asynchronous call

  • The most simple use case is :-

      RxLowpoly.with(context)
          .input(bitmap) 
          .generateAsync()
    

    Along with Bitmaps we can also use Drawables or Files or Uri
    as input.

  • When we need to downscale the image :-

     RxLowpoly.with(context)
         .input(bitmap)
         .overrideScaling(downScalingFactor)
         .generateAsync()
    
  • When we need to set a maximum width for the image :-

     RxLowpoly.with(context)
         .input(bitmap)
         .overrideScaling(maxWidth)
         .generateAsync()
    
  • We can also set a quality for the lowpoly image :-

     RxLowpoly.with(context)
         .input(inputUri)
         .overrideScaling(downScalingFactor)
         .quality(Quality.HIGH)
         .generateAsync()
    

    VERY_HIGH, MEDIUM, LOW and VERY_LOW are also available as
    Quality configurations

  • To save the lowpoly image to a file :-

     RxLowpoly.with(context)
         .input(inputUri)
         .overrideScaling(downScalingFactor)
         .quality(Quality.HIGH)
         .output(outputFile) // An Uri of a File is also supported as an output destination
         .generateAsync()
    

All asynchronous operation is done on the IO scheduler.

Synchronous call

Replacing generateAsync() with generate() in each of the Asynchronous call examples leads to a synchronous call with a lowpoly Bitmap as a result.

A Bitmap of the generated lowpoly image is always returned irrespective of synchronous or asynchronous calls and whether an output File or Uri is supplied using the output method.

Note : A full implementation can be found in the app module of this repository or in the open sourced WallR app.

Critical Analysis

The following tests have been performed on a Xiaomi Redmi Note 5 Pro with 6 gb Ram.

Original Image       Lowpoly Image Input Source Quality Execution Time (ms)
Bitmap Very High 15813
Drawable Very High 16275
File Very High 15987
Uri Very High 15931
Bitmap High 4547
Drawable High 5088
File High 4734
Uri High 4612
Bitmap Medium 1113
Drawable Medium 1672
File Medium 1297
Uri Medium 1152
Bitmap Low 918
Drawable Low 1496
File Low 1091
Uri Low 996
Bitmap Very Low 850
Drawable Very Low 1024
File Very Low 923
Uri Very Low 876

Thus it is evident that when quality is set to High, a good trade-off between speed and texture is obtained, hence the default value of Quality is set to HIGH.

Also, we can see that Bitmap is the input format of choice as it is processed the fastest, followed by Uri, File and Drawable respectively.

Sample App

The sample app provides an implementation of various configurations of RxLowpoly which one can experience and assess first hand before using the library.

Screenshots

       

GitHub

Comments