Frequency-based algorithms are designed to process frequency components of images in frequency domain insted of processing pixles in spatial domain. Therefore, frequency-based algorithms depend on features of images exist in the frequency domain. However, some of those algoerithms can be applied to process image in spatial domain without transforming image from spatial to frequency iff one can compute the effect of Frequncy-based processing on the spatial domain data. I had published the same idea in image watermarking in the following paper:

You can access the paper through this link:

Conference Paper Efficient Mid-band Exchange Coefficient Watermarking System

Long story short: No way (they are muddled )

Let's get to the basics : Long ago when everything was analog , camera starting with lens (operated on a pin-hole concept) were used. The "space " that is light +reflectance (from the subject - (x,y,z)) was transformed to "frequency domain" that is lens would do the "spatio-frequency " transformation and hence an analog signal that has amplitude (Intensity) and frequency (Contrast ) came into being (x,y) !!!!!

Of course you have optical transfer function that gives modulation transfer function (amplitude) (won't go beyond) .

Next, since we already are in the signal domain you can filter /clean up / quantize / etc etc .

Now that you have digital numbers (pixelated ) you can have fun with them . You can transform them (DCT/FFT /IFFT/Convolution ) .

So , here spatial data means (x,y) and intensity variations . While frequency => statistical analysis (Covariance /Mean/Median etc etc)

...

Thanks for the tutoring.