Oooh, a good example of the implications of blown highlights…
When there’s just too much light at a pixel for it to handle, the recorded measurement is just the max number the sensor can produce. Your image has a lot of those, and when the image was captured, all those pixels just maxed out at (more or less) the same number, which could be considered as “white”.
However, that data needs to be whitebalanced, each channel’s data multiplied by a different number to actually make white White. What that does to all of those saturated channels is to shift them away from each other, making the pixel “not-white”. Here’s your image, demosaiced and white balanced, but still linear:
Look at the histogram, and note the separated saturated channels. That’s what makes those pixels “look pink”…
There are a couple of ways to handle that. One is to just set the white point at the green spike, and everything previously pink wouild then go white. I call that “Highlights Abandonment”. The better way to handle it is to use a Highlight Reconstruction operator to use the data between the green and red spikes to make some semblance of definition in those pixels.
Maybe more than you wanted to know… 