# How to distinguish light rays coming from objects of different material

Motivation: My major focus is on Digital Image Processing (specifically segmentation). Due to external noise, the different parts of an image are not fully quantized. Therefore, various segmentation and quantification algorithm has emerged to tackle different types of problems.

For sound wave, there is a property known as 'timber' by which human ear can distinguish sounds (even if they have the same amplitude and frequency) coming from different sources.

Question: Can our eyes distinguish light rays coming from the objects (even if they have the same amplitude and frequency and hence the same pixel intensity when seen through a digital image) of different material?

In analogy to sound, is there any such property of light like timber for sound?

Updated Question (Based on the answers given):

Part1: Consider two real objects of same color and texture but of different materials. Light from a single source fall on those two objects (ignore the difference in distance traveled by the light) and reflected back to the eyes of a person. If the person has normal vision, will he be able to distinguish the objects(assume the person has no prior knowledge about the objects or materials)?

Part2: Consider the same case but now the light is not coming directly from the real objects,rather it is coming from a single digital image of the two objects. Will the same person be able to distinguish the objects now?

Light coming from some source usually consists of many frequencies. Distribution of them is called "spectrum of light". Each frequency (in the visible to a human's eye range) corresponds to some colour. If we limit the spectrum of light to a single frequency, we will see this colour. However, not all colours we perceive can be obtained in this way and also multiple combinations of frequencies will produce the same perceived colour (see e.g. this video and Wikipedia).

In general, a camera will be better at distinguishing the materials by the light they emit/reflect than the human eye. But, other than that, spectrum is the timber of light.

So, if you are able to get the spectrum of light (better yet, if you have also IR frequencies), this is close to a fundamental limit of distinguishing the sources.

### Update 1: Digital images and spectrum

A typical digital image with RGB channels contains a "zipped" representation of the spectrum optimized to maximally agree with human eye (but RGB still can't do many colors). This is implemented as a matrix of sensor triples each having a different sensitivity curve (see here a comparison with human eyes).

So, digital image captures only a part of spectrum that can be represented as a linear combination of 3 sensitivity curves of the sensors. This probably means for you that no, you cannot reproduce the spectrum from a photo. You need a spectrometer for that.

Unfortunately, I cannot tell how accessible this technology is at the moment.

### Update 2: Polarization and computers

Polarization of reflected light also differs some materials, but you need yet another instrument for that (at least, a filter).

Again, human eye and typical digital camera do the same thing: they have a sensor with limited sensitivity range that either lights up or doesn't. The strength of this sensor's signal is all you have. Having 3 kinds of sensors was already cool enough for evolution.

So I would say that from a digital image you can do only as much as you can do with your own naked eye. But then, humans also have computers and algorithms in their heads that take this limited information and by spatial correlations and binocular vision make guesses about the shapes and materials. But this is not exactly physics question at this point.

• Can we reconstruct the spectrum of light from a digital image? Jun 22, 2016 at 11:21
• Updated my answer Jun 22, 2016 at 12:13
• If the spectrum information is not there in the digital image, how is human eye still able to recognize different parts from the image even if this be a grayscale one instead of an RGB? Is there any other characteristic of a digital image apart from intensities, that allow the light reflected from that image provide an object discriminatory information similar to that coming from the real objects? Jun 22, 2016 at 12:30
• Updated my answer Jun 22, 2016 at 12:38

Timbre is a consequence of harmonic content which is a consequence of a sound being made of multiple pure sounds. Thus a 256Hz square wave has pure sinusoidal components at 256, 768, 1280,... Hz.

An equivalent in light terms would be the spectral content of the light. For instance, take the bright yellow D line(s) of sodium on the one hand and a visually identical pigment with a much broader spectral range.

When you detect R, G, B, you are detecting the sum over spectral bands (strictly, a convolution) and not a single frequency. Both yellows might then look the same.

View the scene through a filter, and the colours will look different from each other. This is because the light frequencies which contribute to the R and G signals will not all be absorbed to exactly the same extent as the sodium D line.

Many eyes in the animal kingdom use filters to provide a wider variety of colour signals without needing too many different colour receptors: hence 4-colour or even 5-colour vision.

Example

Consider the green sensor. A signal at 550nm will contribute twice as much as one at either 510nm or 580nm, and you have no way of distinguishing between "standard 550nm" and "double-brightness 510nm" or "double-brightness 580nm". The result will be precisely the same, and indistinguishable.

But suppose that you put a filter in front of the lens which passes 50% of the light at 510nm, 75% at 550nm, and 100% at 580nm. Then if the pixel doesn't change when you add the filter, you will know the light was at 580nm. If the pixel gets a little dimmer, it's more like 550nm. If it gets much dimmer, then it is towards 510nm. Technically, you are convolving the original spectrum not just with the sensitivity of the G sensor but also with the transmittance of the filter.

Of course, in reality, filters aren't so simple, but this serves as an example.

So with two photographs, instead of just "green", you can identify "yellowish green" and "bluish green". In principle you could extend this with more filters; or if you had a filter disc with different transmittances rotating in front of the camera lens, you could get a time-varying signal whose amplitude and phase told you a lot about the "timbre" of the colour.

• You could thus take several photographs of the same scene through different filters and thus gain more detail than the simple R, G and B channels. You could even (if the application permits) have a spinning colour filter in front of the lens, in which case "timbre" differences would be apparent as a time-dependent signal. Jun 22, 2016 at 10:44
• Are the original spectral bands still preserved in the digital image? Or, can we reconstruct the bands? Jun 22, 2016 at 11:18
• No; and yes, to some extent. See my edit. Jun 22, 2016 at 17:48

Does light carry any information from the source?

Definitely. Light has properties that can be used to determine material of the source, as well as material through which the light ray has passed.

Spectrum of light, is a great source of information which is actually equivalent to timber for sound. This is how astronomers find out what elements make up a particular star.

Polarization of the light is another property, and this one doesn't exist in sound because of the basic difference between longitudinal and transverse waves.

Can human eye/digital camera capture any of this information?

Almost no. Human eye isn't an spectroscopy instrument. It captures something like a shadow from the actual light spectra. Imagine a grayscale photo. Can you find out the exact colors? Most of the time you can't, but you can guess like this is not black or that's not white.