It depends. If you have cloud-free clear images with flat terrain, atmospheric correction should not bring much improvement to classification results based on clustering algorithms. If unsure, just try with and without.

In any case, no improvement would result from simple techniques as turning DN into TOA radiance or reflectance or even the simplest DOS atmospheric correction, which just subtracts dark object value from the whole image, if you just use the bands directly for classification after such transformation. That is because all these methods are just linear transforms of the whole bands and these do not have influence on clustering algorithms.

A more sophisticated method of atmospheric correction, which takes into account different atmospheric properties in different parts of the image effectively removing the differences not caused by surface properties between these parts may improve classification results a lot, especially in images, which are more affected by atmosphere.

In complex terrain a topographic correction or a transformation of the bands suppressing the influence of slope lighting may bring additional enhancement (and for that the before mentioned simple techniques may be necessary).

Keep in mind also, that atmospheric or topographic correction would bring improvement only in case it is done right. It can happen atmospheric correction "over corrects" image in parts because of shortcomings in AOT retrieval for example, or topographic correction brings artifacts all over the image because of DEM deficiencies.

this PdF may be help

If you are solely interested in Landuse classification you don't as far as you can still classify those landuses. That is the hortest answer

Yes, you should. however most of GIS professionals normally skip this step because normally it does not affect so much your work [...]

Therefore, once you have your data in radiance, it still have the effect of the light source, atmosphere, surface material, and typically we are interested in the surface materials. So, that why Atmospheric Correction (AC) comes into pictures. once you performed AC on the data that's already has been calibrated to radiance, removed the effect the effect of light source and atmosphere, we will end up with the information about just the surface material. Actually is very easy to carry it out, you should read the the metadata file, understand the formula explained in landsat project, and then performe it in raster calculator. All the information that you need is in the metadata.

Hope this could help you...

It depends. If you have cloud-free clear images with flat terrain, atmospheric correction should not bring much improvement to classification results based on clustering algorithms. If unsure, just try with and without.

In any case, no improvement would result from simple techniques as turning DN into TOA radiance or reflectance or even the simplest DOS atmospheric correction, which just subtracts dark object value from the whole image, if you just use the bands directly for classification after such transformation. That is because all these methods are just linear transforms of the whole bands and these do not have influence on clustering algorithms.

A more sophisticated method of atmospheric correction, which takes into account different atmospheric properties in different parts of the image effectively removing the differences not caused by surface properties between these parts may improve classification results a lot, especially in images, which are more affected by atmosphere.

In complex terrain a topographic correction or a transformation of the bands suppressing the influence of slope lighting may bring additional enhancement (and for that the before mentioned simple techniques may be necessary).

Keep in mind also, that atmospheric or topographic correction would bring improvement only in case it is done right. It can happen atmospheric correction "over corrects" image in parts because of shortcomings in AOT retrieval for example, or topographic correction brings artifacts all over the image because of DEM deficiencies.

I agree with what Tomáš Brunclík said. To apply atmospheric correction in ENVI you can execute these steps

1) open your image in ENVI using metatdata file

2) execute the function "calibration image" to convert DN to radiance image

3) execute FLAASH algorithm to convert radiance image to reflectance image

due to some errors which can accur during execution of FLAASH, the reflectance image can have négative values or value exeding 10 000. to correct it you have to bound the values between 1 to 10 000 using this expression

(b1 le 0)*0 + (b1 ge 10000)*1 + (b1 gt 0 and b1 lt 10000)*float(b1)/10000

FLAASH work generally well with landsat 4 to 7, but for L8 it can result with a bad image for near infrared band.

Lets try and see

Thank you so much for your suggestions. Reading all the suggestions, I would like to compare the images with and without the Atmosphere correction.

I have tried the FLAASH with ENVI for Landsat 8. It does not seems any difference in terms of visualization.Moreover, I tried to take that image in Erdas Imagine as a file format .img but it could not load. The layers are missing. Could anyone explain this? In addition, I was trying different methods of atmospheric correction to improve the image quality to identify the objects, I tried 'Internal Average Relative Reflectance (IAR) in Erdas for landsat 8 and the output image is much better to identify the objects. Could anyone explain me this?

Follow this file how to Atmospheric Correction for the Landsat images

And also follow this video get more information on how to process the Atmospheric Correction for the Landsat images

I agree with Tomáš Brunclík and Cédric Aurelien Nguimdo Matsaguim