To substantiate earlier answers on perception of the true color image, one complement the analysis with image statistics as a tool less prone to personal bias.
This attempts a synthesis of the contribution by the answers by @Tetsujin (because of the provision of a prior/after processing illustration) in comparison to the one of the squirrel (answer by @Danielillo) and the preview of your question.
Art aside, photography equally is a measurement; and every measurement has some random error. Inconsistencies in processing (electronic) photos can yield a non-uniform error level across a JPEG image which as such can be visualized like a map. One of the programs accessible to the public I'm aware of is forensically by Jonas Wagner created around 2012 (blog post, and 12 min youtube tutorial; both in English).
Let's submit the two versions of the bird image by @Tetsujin to perform an Error Level Analysis. To highlight the differences, an alternating gif of the two annotated illustrations:
(analysis based on the .png files shared, vide infra)
The differences might be subtle: watch the areas of patches of similar (false) color which tend to be more uniform for the smoothed illustration. For training purpose the same was applied on the portrait of the squirrel. We observe two level of noise here: the background, on one hand vs. animal and tree on the other. This is easy to comprehend by the use of a tele lens; here, used to «get close enough» (and in classical portrait photography, to separate the object from the less interesting background which should not distract too much from the former).
(based on the .png shared in the answer)
(based on the .jpg from the author's web page here)
The comparison shows that the file format can influence the result of this analysis (after all, .png are more suitable for screen photos anyway).
And eventually, the albatross. Bodies and necks, but not (so much) the feet clearly stand out this much in comparison to the surrounding grass that a high level of editing is likely. And the sky in the background.
(based on the .jpg of the preview here).
The similar tool fakeimagedetector offers a slider to show the true color image/false color representation of the noise map. After some time, you can't unsee the variation in coarseness in the later.
(screenphoto from https://www.fakeimagedetector.com/)
There are multiple types of errors which add up to noise in the image.
- photon shot noise: light is both wave and particle, but the arrival of photons on the detector is noncontinuous and random. The emission of photons need not be homogeneous across a surface area inspected, either.
- electronic noise of the readout: recording an image is assigning a pixel an intensity value. To obtain a larger number of charges per pixel eventually easier to process, this amplification may introduce an error.(1)
- quantization noise of the readout: the (amplified) analog signal is converted into a digital one. (Your analog signal may end up in the wrong digital bin.)
Aside from these, the pixels within a camera chip are not uniform; right from production, some may be less sensitive, than others. Repeated overexposure/too high levels of gain (especially ICCD camera), etc. may lead to dead pixels no longer recording a signal. Second, a pixel in operation can report intensity even in absence of a radiation source adding dark current noise. This however is more relevant to intentional observation of dim scenes / experiments with an
extended period of exposure (e.g., fluorescence microscopy, astrophotography, X-ray crystallography) where one intentionally awaits temperature stabilization of the area detector, and collects dark frames for a subsequent background collection on individual frames. See (3) the above paraphrases.
These influences are all ahead of storing the information as .csv, .pgm, plain/LZW compressed .tiff. Or before the camera equally corrects white balance,
adjusts sharpness, starts other electronic/AI filters (modern cellular phones) and applies a lossy compression like JPEG for the file available for you.(4) Subsequent edits on JPEGs require to encode parts of the retouched areas which then distort the file further, this and the transitions is spot by the program.
(1) With ICCD cameras,(2) this gain actually can be adjusted and can introduce additional problems down the road (leaving the linear range / where recorded signal is proportional to the intensity).
(2) Without particular endorsement, e.g. LaVision.
(3) Youtube videos Microscopy: Cameras and Detectors I: How Do They Work? and Microscopy: Cameras and Detectors II: Specifications and Performance by Nico Stuurman as
part of a whole class around light microscopy and relevant image processing.
(4) How are Images Compressed? by BranchEducation is an instructive video about the steps of JPEG compression (color space conversion, chrominance downsampling, discrete coside transform, quantization the eventual Huffman encoding). It both explains the steps, and one gets an idea why recording a photo in a raw format rather than JPEG can be advantageous. (Not every image format constrains the the intensity depth to 8 bit per (color) channel (i.e., 256 levels) only.)