This question has essentially been asked before:
In short, resizing is a geometric transformation, or more precisely a scale transformation. For each pixel in the output image, imresize is doing a point transformation to find where the corresponding point maps in the input image. Once imresize has formed a collection of remapped points, interpolatation is performed to resolve sub-pixel remapped locations (the default of imresize is bicubic interpolation).
If the output image is being scaled with a scale factor less than 1, then by default a low-pass anti-aliasing filter is applied to the input image prior to the steps described above to prevent high frequency content from the input image introducing distortions due to downsampling the input image. You can turn this anti-aliasing filtering off to see how it effects results. It's generally a very important step when down sizing an image to get visually nice looking resized images.
