You will add the functionality to a skeleton version of the Impressionist program, which we will provide. The purpose of this project is to give you experience working with image manipulation, OpenGL primitives, user-interface design, and image processing.
Impressionist is an interactive program that creates pictures that look like impressionistic paintings. It is based on a paper and a program by Paul Haeberli. Here is a copy of his paper "Paint by Numbers".
To create an impressionistic picture, the user loads an existing image and paints a seqence of "brush strokes" onto a blank pixel canvas. These brush strokes pick up color from the original image, giving the look of a painting. See some examples of what you can do with Impressionist, from previous quarters.
This is motivated by impressionist paintings that artists have been making for centuries. To help inspire you when creating new brush types, or when creating your artifacts, here are the Google Images "Impressionist Art" results.
All of the skeleton code for the projects is stored on gitlab. We strongly recommend using git (or another) version control system when working together with your partner.
Visit here for help checking out code and information on setting up and using git.
The skeleton program we provide does very little. It allows you to load the original image (which can be in BMP, PNG, or JPEG format), and save the painted version. Brush selection is done via a drop down list on a separate window called up via the "File" menu. There is one brush implemented (points) and a slider for controlling the brush size.
The skeleton comprises the following classes. The following descriptions should give you a good idea of the purpose of each of the classes and provide some insight into where to add things when extending the skeleton code.
This is where the user interface for the Impressionist project is defined. Add new widgets to your UI here.
This basically maintains the bitmap for answering the color query from the brush. +
It also acts as the bridge between the brushes and the UI (including views).
This is a base class for each of the paint brushes.
It defines the functionality that the brushes should have.
Your new brushes should inherit from this class.
The color that your brush paints with is also set here.
This maintains the paint view of the input images and handles events related to the paint view.
Add new events and changes to the painted image here.
This maintains the original view of the input images. Add changes to the original image here.
This is the implementation of Point Brush. It is a kind of ImpBrush. All your brush implementations will look like this with different GL primitive calls.
Implement 5 different brush types:
single line, scattered lines, scattered points, (filled) circles, and scattered (filled) circles.
See the sample solution for an example of each brush's appearance.
Note that scattered brushes should sample from each location they color individually,
not just use a single color for each splotch.
Add sliders to control various line brush attributes.
You need to include sliders for the thickness (width) and orientation (angle) of line brushes,
in addition to the existing brush size slider.
Add the ability to control the brush direction.
The line brush orientation should be controlled four different ways:
using a slider value (see above), using the right mouse button to drag out a direction line,
using the direction of the cursor movement, and using directions that are perpendicular to the gradient of the image.
You can use a radio box to allow the user to select which method to use.
Use the Sobel operator to determine the gradient.
The immediate brush directions obtained from mouse movement can be quite noisy,
and it is nicer if they are smoothed out with respect to previous brush directions.
You will need to handle this issue. Note that, if you remember the brush directions as a list of angles,
you cannot just average the angles. A simple average doesn't account for the fact that two brush angles,
such as 359 degrees and 0 degrees, may be closer than their numeric values imply.
We recommend using the atan2(y,x) function instead of atan(y/x),
because it handles the case where x=0. Note that its return value is between -pi and pi.
Allow the user to change the opacity (alpha value) of the brush stroke.
An alpha value slider should be added to the controls window.
You may want to consult the blending subsection of the OpenGL Red Book for this. (Pages 162-169 of the linked PDF are particularly helpful.)
Make sure that you call glBlendFunc *outside* of glBegin/glEnd.
Only a subset of GL commands can be used between glBegin and glEnd, and glBlendFunc is not one of them.
Implement the filter kernel.
The skeleton code already provides the user interface.
You should be able to specify any 5x5 filter by typing in the filter coefficients,
a scale factor by which each filter coefficient is divided,
and an offset which is added to the pixel before displaying.
The filter is applied to the entire "Paintview" image (the image on the right panel).
You will need to implement a method for handling boundary pixels,
when part of the filter kernel goes off the edge of the image.
You must do something "smarter" than assuming that the image is black beyond its boundary (i.e., do something other than zero-padding).
The filter kernel allows users to enter values such that the resulting pixel value is out of the range [0...255] or the kernel values are divided by 0.
You will need to do something "reasonable" to handle these cases.
The filter is applied to the painted image (instead of the original image).
Also, the filter kernel dialog is a modal dialog.
You must first close the filter kernel dialog before you can continue painting on the "Paintview" canvas.
Also, be advised that OpenGL stores the image in memory upside-down - i.e if the image is a 3 x 3 grid with [1 2 3] as the top row followed by [4 5 6] and [7 8 9] as the other two rows,
then the array will be [7 8 9 4 5 6 1 2 3]. You will need to compensate for this when implementing the filter kernel.
Implement at least one bell's worth of bells and whistles.
See below for the list of bells and whistle suggestions. Any additional will be extra credit.
1 Bell = 2 Whistles
Please follow the general instructions here. More details below:
For the Impressionist artifact, you and you partner will each create an impressionistic painting from an image of your choice. Please turn in both the original and impressionized version in jpg or png form.
Bells and Whistles
Bells and whistles are extra extensions that are not required, and will be worth extra credit. You are also encouraged to come up with your own extensions for the project. Run your ideas by the TAs or Instructor, and we'll let you know if you'll be awarded extra credit for them. If you do decide to do something out of the ordinary (that is not listed here), be sure to mention it in a readme.txt when you submit the project.
To give your paintings more variety, add some additional brush types to the program.
These brush strokes should be substantially different from those you are required to implement.
You will get one whistle for each new brush (within reason).
The skeleton program allows the user to paint outside the boundary of the paint rectangle,
then erases this region when the stroke is completed.
Change this to clip brush strokes to the region as they're being painted.
When using your program, you currently can't see what part of the original image you're painting.
Extend the program so that when the cursor is in the painting window,
a marker appears on the original image showing where you're painting.
Sometimes it is useful to use the contents of the painting window as the original image.
Add a control to swap the contents of the painting window and the contents of the original image window.
For your filter kernel, add a "normalize" checkbox that will automatically divide by the sum of the weights when the user wishes it.
Add controls that allow you to manipulate the color of the image.
For example, you could implement independent scaling of the red, green, and blue channels.
Design a brush that selectively applies one or more filters from your filter kernel.
This might require some UI changes to your filter kernel UI.
Note: you must take into account the brush size.
Add an undo feature with at least one level of undo so that you can try a brush and decide to undo its effect on the canvas.
This comes in very handy for experimenting with brush and filtering effects.
Add the ability to dissolve one image into another.
Fix the Red Line Bug in right mouse drag out direction line.
It is possible for a series of odd mouse clicks during a right mouse drag to permanently draw the red line on the canvas (spoiling a possibly beautiful drawing).
Eliminate this problem entirely by properly synchronizing mouse events.
A different solution to the problem of not being able to see where you're painting is to show a dimmed version of the painting on the canvas.
Add a slider that allows the user to fade in or fade out the original image beneath the user's brush strokes on the canvas.
(Beware, this bell and whistle is more difficult than it looks).
Add a "mural" effect to your Impressionist by implementing the ability to load in different images while preserving what has been drawn on the canvas.
Add a "New Mural Image" or "Change Mural Image" to the controls window that allows the user to change images.
The user may then load an image, draw in what he / she prefers on the canvas,
and then load a different image and continue drawing on the canvas; thus, a "mural" effect.
To make your painting more interesting, add "alpha-mapped" brush strokes.
In other words, allow the user to load a bitmap representing a brush stroke.
This bitmap would contain an alpha value at each position.
Then when this brush is used to draw, a single color would be selected from the image,
all pixels in the brush bitmap would be set to this RGB color (without changing the alpha value),
and this partially transparent bitmap would be painted on the canvas.
A new color would be used each time the brush is drawn.
It can be time-consuming to paint an image manually.
Add a feature so that a whole painting can be created automatically.
The user should only have to specify a brush type, size, and angle to use.
Then the program should automatically paint brush strokes over the entire image,
using a randomized brush order and varying the brush attributes slightly as it goes (to increase realism).
At times, you may want the brush strokes to follow the gradient of a different image than the base image.
Add a button(s) that will cause the direction of brush strokes to be automatically determined from a user specified image.
The "accuracy" of the painting can also be improved by clipping long brush strokes to edges in the image.
Allow the user to load a black-and-white image that represents the edges in the picture.
Then add a checkbox so that the user can turn on edge-clipping,
which will automatically clip brush strokes at edges in the image.
Use the image processing techniques described in class to automatically find the edges in the base image.
Once you have found the edges, add a button to the user interface that will allow the user to select whether or not the brush strokes should be clipped to the edges in the picture.
Implement a bilateral filter. You should provide sliders that control how wide the filter is: one slider for the domain width,
a second slider for the range width. You earn one bell for implementing this filter just as a specialized mean filter,
where you simply average values that are within the domain/range widths of each image pixel. For an extra bell,
use Gaussian filters, and add a toggle to switch between bilateral mean and Gaussian filtering.
Design a brush that can be used to stretch and pull the image as if it were rubber.
Implement "animated" brush strokes that make the image appear to move in interesting ways.
Ex. You could paint moving ripples over a picture of a lake, or rustling motions onto grass or tree.
Credit will vary depending on the success of your method.
Given a source image, construct a new image that is really a mosaic of small (thumbnail) images.
To do this, you need to partition the original into tiles and find new thumbnails that are reasonable matches to the tiles.
Then draw the new image by substituting the thumbnails for the tiles. See, for example,
Adam Finkelsteins Web Gothic.
Mosaic sample solution (stand-alone command line executable).
Credit will vary depending on the success of your method.
To get full credit, you must perform some sort of edge detection to accurately determine which thumbnails to use,
and you must use the original color of the selected thumbnails.
Disclaimer: please consult the course staff before spending any serious time on these. These are all quite difficult (I would say monstrous) and may qualify as impossible to finish in the given time. But they're cool.
Implement a method to automatically create non-photorealistic video. One very simple method (that would not get a monster bell) would be to run auto-paint on each frame of a sequence. For credit, your technique should exhibit temporal coherence.
Other artistic methods, such as charcoal sketch, often de-emphasize the background (or leave it out altogheter). When processing still images, it is practically impossible to distinguish the subject and the background without any human assistance; however, in a video stream, it may be possible to exploit movement to segment the image. For additional credit, implement a method that effectively uses this method to generate a convincing non-photorealistic version of live video.
For even more extra credit (and probably a conference paper) do all of this in real-time on a consumer PC.
Image mosaics are often pieced together by stitching together a bunch of tiny rectangular images. Although this produces a cool effect, it looks computer generated. Implement a method to build collages, given a sample set of images. The primary difference is that the shapes need not be rectangular and that they can also overlap. A while back, some graduate students here implemented a method to do this, ultimately resulting in building a face with pictures of fruit.
Another approach is to note that, when humans build collages, we usually clip shapes out of images (cutting out a picture of a red car and pasting it in as someone's upper lip, for instance). Given a set of data images, we wish to automatically build a collage of some input image, given that we can cut simple shapes from the data images. If you've seen The Truman Show, you may remember that Truman puts together a picture of a woman's face using magazine clippings. This took him a while.
Okay, I completely made up that term. In artistic animations, the movement is often not completely realistic. One technique that has been used for some advertisements and music videos involves sampling the video at a very slow frame rate (say, two per second) and then filling in the disarded frames using morphing. You may want to use the optical flow (see CSE490CV) to assist with the morph. You may also want to split up the image, morphing different regions and varying frame rate according to how much movement there is. After you perform this pass, maybe try running your impressionist program on each frame, using the morph and the optical flow to guide the direction of temporally coherent brush strokes.
For even more extra credit (and probably a conference paper) do all of this in real-time on a consumer PC.
Frequently Asked Questions
What if the sample solution doesn't run on my computer?
If you have run the Impressionist sample solution without VS2015 installed on your machine, you may encounter an error about incorrect application configuration. The reason is that this executable requires some VS2015 DLLs. You can get past the problem by either installing VS2015, or installing just the DLL's from here. Please note this solution only works for release mode executables (which the sample solution is).
Why doesen't glEnable work inbetween glBegin and glEnd?
This is because the glBegin-glEnd block works with the current state of the computer.
Is there a way to catch OpenGL errors?
Yes. When it seems like nothing is happening, OpenGL is often returning an error message. To check for this, use the following code: