Font segmentation

coolbutuseless-arcadefont.R

library(magick)
library(hocr)
library(bunny)
library(bbx)
library(dplyr)

img <- image_read("https://pbs.twimg.com/media/EXKzsiAWoAYH9Pv?format=png")

img_bw <- img %>% 
  image_threshold("white", threshold = "5%") %>% 
  image_convert(type = "Grayscale") %>% 
  image_morphology("Open") %>% 
  image_negate() 

bbox_df <- img_bw %>% 
  image_ocr(HOCR = TRUE) %>% 
  hocr_parse() %>% 
  tidy_tesseract()
  
bb <- bbox_df %>% 
  slice(1) %>% 
  pull(ocrx_word_bbox)
  
img_ltr_slots<- image_crop(img_bw, geometry = bbx_to_geometry(bb)) %>% 
  image_morphology("HMT", paste0("Rectangle:1x",bbx_h(bb))) %>% 
  image_morphology("Thinning", "ThinSe")

## hough line method
ii <- image_info(img_ltr_slots)
iiw <- ii$width
iih <- ii$height

img_ltr_slots %>% 
  image_hough_txt(geometry=paste0("1x1+", iih)) %>% 
  bunny::tidy_hough_mvg() %>% 
  .[["lines_data"]] %>% 
  filter(line_angle==0) %>% 
  pull(line_bbox) %>% 
  bbx_to_bbm() %>% 
  {rbind(matrix(c(0,0,0,0), nrow = 1),.,
         matrix(c(iiw, iih, iiw, iih), nrow=1))} %>% 
  {mapply(function(i, f) f(.[,i],-1), seq(ncol(.)),
       list(head, head, tail, tail))} %>% 
  bbm_to_bbx() %>% 
  bbx_translate(bbx_to_geometry(bb)) %>% 
  bbx_pad_width(side="right") %>% 
  bbx_to_geometry() %>% 
  lapply(function(x)image_crop(img_bw, x)) %>% 
  {Reduce(c,.)} %>% 
  image_scale("300%") %>% 
  image_animate(delay=50) %>% 
  image_write_gif("coolbutuseless_arcadefont.gif")