After finding a typographic portrait of Christopher Hitchens, created out of Gill Sans letters by Miles Chic at Capilano University, I thought to resurrect software I wrote a long time ago that converts images into letters and expanding traditional ASCII art by using
The representation of images by characters—ASCII art—has a long history. ASCII art extends the emoticon (or smiley) to represent a larger piece of work. Typically, the works use a fixed-space font (e.g. Courier), originally designed for display on a terminal. Despite the sophistication of computer graphics today, ASCII art continues to have a strong following with new work continually added to public online galleries.
Photos and paintings can be ASCIIfied using a tone-based approach and automated methods exist to do this (Paul D. O’Grady and Scott T. Rickard (2008) Automatic ASCII Art Conversion of Binary Images Using Non-Negative Constraints).
Many artists generate new creations, exclusive to the medium. Typically this kind of ASCII art is based on the interpretation of structure rather than tone—this method has also been automated (Xuemiao Xu, Linling Zhang, Tien-Tsin Wong (2010) Structure-based ASCII Art).
I have written code to generate ASCII art from images by using proportional spaced fonts.
Let's see how these methods work on a real image. Many ASCII art Mona Lisa versions exist. Below, I render the Mona Lisa with Pragmata, Gotham Book and 8 weights of Gotham.
Two-tone shapes like the S in the figure above require selecting characters that match the structure of the image. (e.g. "|" matches vertical lines). For a given character and image position there are four distinct match possibilities—a combination of whether the character and image have a signal at a position. I show this in the figure below.
By maximizing scores derived from matches (s1, s3) and minimizing any penalties (s2, s4), a character is identified based on maximal coverage of the image region and minimum coverage of areas that are blank.
When proportional text is used, edges are better approximated, such as in the Homer Simpson example below which uses Gotham Book.
Images that are not two-tone require that we match both structure and tone. Structure is approximated by the choice of character, while tone by choice of font weight. To select the best character based on tone, the character's average tone is compared to the average tone of the section of the image to which it is being compared.
It is possible to combine both structure and tone metrics in character selection. Below is an example of how an image with both tone and structure is interpreted as the tone and structure score weights are varied. The balance between these two metrics can be very hard to find—it greatly depends on the image. Tone-based mapping works well when font size is small and the image is viewed from larger distance—in this case, characters play the role of individual pixels with varying brightness. Structure-based mapping works with larger type and closer viewing distance.
Continuous tone bitmaps are an idea application of multi-font ASCII art—images no longer need to be thresholded or dithered.
ASCII art is generated by dividing the image into a grid and finding the letter (the choice of characters is often expanded to include punctuation) that best matches the grid section. Typically, for each grid the entire set of allowable characters is sampled. Instead, we can limit the choice of character by successively sampling from a fixed string.
rendered with the fixed string "monalisa" using 8 weights of Gotham.
Things get even more interesting when the text is angled.
The image can be textured with multiple layers of ASCII art. In the example below, four layers of text are used, each with a different font size.
Instead of varying size, the angle of the text can be changed among layers. This results in a pattern reminiscent of a halftone.
An image can be asciified several times, with each iteration the asciified output of the previous step used as input for the next. At each step, the font size should be reduced to s → √s.
In this primer, we focus on essential ML principles— a modeling strategy to let the data speak for themselves, to the extent possible.
The benefits of ML arise from its use of a large number of tuning parameters or weights, which control the algorithm’s complexity and are estimated from the data using numerical optimization. Often ML algorithms are motivated by heuristics such as models of interacting neurons or natural evolution—even if the underlying mechanism of the biological system being studied is substantially different. The utility of ML algorithms is typically assessed empirically by how well extracted patterns generalize to new observations.
We present a data scenario in which we fit to a model with 5 predictors using polynomials and show what to expect from ML when noise and sample size vary. We also demonstrate the consequences of excluding an important predictor or including a spurious one.
Bzdok, D., Krzywinski, M. & Altman, N. (2017) Points of Significance: Machine learning: a primer. Nature Methods 14:1119–1120.",
Just in time for the season, I've simulated a snow-pile of snowflakes based on the Gravner-Griffeath model.
Gravner, J. & Griffeath, D. (2007) Modeling Snow Crystal Growth II: A mesoscopic lattice map with plausible dynamics.
We introduce two common ensemble methods: bagging and random forests. Both of these methods repeat a statistical analysis on a bootstrap sample to improve the accuracy of the predictor. Our column shows these methods as applied to Classification and Regression Trees.
For example, we can sample the space of values more finely when using bagging with regression trees because each sample has potentially different boundaries at which the tree splits.
Random forests generate a large number of trees by not only generating bootstrap samples but also randomly choosing which predictor variables are considered at each split in the tree.
Krzywinski, M. & Altman, N. (2017) Points of Significance: Ensemble methods: bagging and random forests. Nature Methods 14:933–934.
Krzywinski, M. & Altman, N. (2017) Points of Significance: Classification and regression trees. Nature Methods 14:757–758.