When resizing images in Adobe Photoshop, you have the choice of three resampling modes. Below is the dialog box that shows you the options in the dropdown.

The table below shows a sample for resizing in each mode.

	Nearest Neighbor	Bilinear	Bicubic
50%
66%
100% Original image
133%

If you think about an image as a bunch of pixels (or dots), and you want to reduce and image by 50%, you would intuitively just remove every other pixel. This would be OK, but what about the small details in your original image that are only a single pixel wide? They would disappear. This is where bicubic interpolation comes in. It does a kind of "averaging" of the colors when deciding what pixels to remove.

Now what about when you resize an image by something that is not a power of 2 (i.e. 2, 4, 8, etc.) What pixels do you get rid of? You can see the affect of this above in the 66% row. You can see in the nearest neighbor column, the image is very bad. The bicubic image is more representative of the original. Also remember that this is a rather extreme sample image to work with and is only used to show you the difference between the resampling modes. Also note that the bicubic image is a little "fuzzy". This is how Adobe Photoshop handles the problems of getting rid of an odd number of pixels across an image. The difference in "averaged" over the entire image rather than just ignored.

Below is a sample of a photo resized from 1760w x 1168h pixels to 512w x 340h pixels. Notice the window shutters on the left of the image, the siding, and the bricks. All look a little "funny" in the first image. What you are seeing are artifacts produced by aliasing. Just think of it as getting rid of pixels without being very smart. Using "bicubic" resampling, you are telling Adobe Photoshop to be a little more smart in how it resizes and image. Years ago, resampling using bicubic interpolation was rather slow. Now it is the default mode when you resize an image, and for a good reason.

Nearest Neighbor
Bilinear
Bicubic