Use morphological operations to turn a pixel based image of text into geometric information (x,y locations), sizes, etc. to allow direct comparison
I used rstats, RStudio and some cran and Bioconductor libraries. Check code.
bwlabel)starting out with this
and get this
and then get this
| # label and position information from text images | |
| # | |
| # ds 2020-08-18 | |
| library(imager) | |
| # BiocManager::install("EBImage") | |
| library(EBImage) | |
| library(tidyverse) | |
| # load image -------------------------------------------------------------- | |
| fname <- '1.png' | |
| textImage <- EBImage::readImage(fname) | |
| # and make sure it's grayscale | |
| textImage <- channel(textImage, "gray") | |
| # display | |
| display(textImage) | |
| # threshold | |
| textImage = thresh(textImage, 10, 10, 0.05) | |
| display(textImage) | |
| # fill / erode / turn to blobs -------------------------------------------- | |
| textImage_blobbed <- closing(textImage, makeBrush(7, shape='disc')) | |
| textImage_lablelled <- bwlabel(textImage_blobbed) | |
| display(textImage_blobbed) | |
| display(textImage_lablelled) | |
| # compute + convert to dataframe | |
| fts <- computeFeatures.shape(textImage_lablelled ) %>% | |
| as_tibble() | |
| # some useful metrics | |
| glimpse(fts) | |
| # the moments are central tendency of x, y of each labelled blob | |
| # major axis, angle, etc. (assuming bivariate gaussian, I assume - check doc'n) | |
| ftm <- computeFeatures.moment(textImage_lablelled ) %>% | |
| as_tibble() | |
| glimpse(ftm) | |
| # now plot moment data on image ------------------------------------------- | |
| d <- dim(textImage) | |
| ftm %>% | |
| ggplot(aes(x = m.cx, | |
| y = m.cy, | |
| angle = m.theta, | |
| radius=m.majoraxis)) + | |
| geom_spoke() + # maybe not? not quite correct /// but to see what you can do | |
| geom_point(size=2, color="red", alpha=0.5) + | |
| scale_y_reverse(limits=c(d[2],0)) + | |
| scale_x_continuous(limits=c(0, d[1])) + | |
| labs(x="pixels in x", y = "pixels in y", title = "centroids and major axis / theta") + | |
| theme_minimal() | |
