The first technique we’ll consider is that of replacing, wholesale, the entire background region of the image.  In section 12.6 we considered methods for de-emphasizing the background, particularly through the use of massive blurring to achieve an effect similar to the smooth bokeh from a shallow depth of field.  Two limitations of that method are that the resulting backgrounds will still have roughly the same color as before (though it can be modified via hue and saturation adjustments), and that the resulting backgrounds can end up so smooth and textureless that they look boring and artificial.  While it’s generally desirable to keep the background from being too detailed and distracting, moderate amounts of detail in the background can be beneficial, because it gives the viewer’s eye something to contemplate after s/he has had a chance to take in the foreground.  Indeed, as we mentioned in section 8.1, the visual dynamics that result from foreground and background competition for the viewer’s attention can (in some cases) improve the overall aesthetics of an image.  When the original background of an image doesn’t satisfy these considerations, replacing the background with that from another photo can sometimes help to improve the image.
    The figure below illustrates some of these subtle issues.  The image on the left is the original; the image on the right shows the result of replacing the background with a less “busy” one.  The image on the right clearly focuses the viewer’s attention on the bird more than the image on the left, but it also drastically alters the overall color composition of the scene.  In cases like this, it may be better to instead explore your options for improving the existing background through the judicious use of blurring and selective editing (e.g., removing the tree branch in the background via the Clone tool—see section 11.5).



     Actually replacing the background is, in many cases, not nearly as difficult as you might imagine—as long as you’ve got a background image that you can use.  You’ll generally want to use an entire image for the new background, rather than trying to copy the background from part of another bird photo.  I make a point of taking a few severely out-of-focus photos (of the scenery, not of a bird) in each shooting environment I find myself in, in case I need to replace any bird photos’ backgrounds and want the color composition to remain consistent.  I also try to collect such background images into a central place in my archives, so I can easily find a background when needed.
    The following series of figures illustrates the actual replacement technique in simple steps.  The first step is to select the foreground—including both the bird and any other elements that you want to retain in the image—and to save that selection into a named channel (see section 10.6).  The figure below shows the original image and foreground selection.  Note that for this application (replacing the background) it’s typically the case that you’ll want to be fairly exacting in your selection; for this example I zoomed in and refined the edges of the selection by nudging them as needed with the Quick Selection tool.  As we’ll see in a bit, errors in the selection can be mended later by instead modifying the layer mask that we’ll be creating for the background layer; nevertheless, I personally prefer to refine the selection boundaries during the actual selection phase, since I find the Quick Selection tool to be very efficient for this task, even when working at the level of fine details on the zoomed-in image.  Just how finely those details need to be followed depends, of course, on what you intend to do with the image: for posting small resolution images on the internet, the finest details can often be ignored, while for printing on large media you’ll likely want to be more exacting.



     Once you’ve got the foreground selected, you can invert it (via Select > Inverse) and then save the resulting background selection to a named channel.  You can now load your background image into Photoshop, select the entire background image, copy it via Cmd-C / Ctrl-C, and then paste it as a new layer in the first file (via Cmd-V / Ctrl-V).  In the Layers panel you’ll see two layers: the original foreground layer at the bottom (which will, unfortunately, be automatically labeled “Background” by Photoshop), and the new background layer, which will be above it.  Next, click on the new background layer in the Layers panel, and bring up the background selection via Select > Load Selection (or, alternatively, bring up the foreground selection and invert it).  The inverted outline of the foreground should now be superimposed on the background image, via the original foreground selection.  With the selection still active, click on the Add Layer Mask button at the bottom of the Layers panel.  A layer mask will be created, with the mask initialized to the (inverted) shape of the foreground.  The figure below illustrates what you should see in your Layers panel after doing this.

     In the figure above, the bottom layer is the original image, the green layer above it is the photo to be used as a new background, and the black-and-white box next to the green layer is the layer mask for the new background.  Note that the layer mask is shaped like a bird perched on a branch.  This shape came from the (inverted) foreground selection that we crafted from the original image.  What the layer mask does is to make certain parts of the affected layer transparent.  In this case, the bird and the branch it’s perched on will now show through the top layer from the layer below, because those regions are masked out (via the layer mask) of the top layer.  The result is shown below.



     Notice in the figure above that the far end of the branch looks odd: its edges meet too abruptly with the background.  The problem is that the far end of the branch is out of focus (due to the shallow depth of field) and therefore soft, but the edges appear unnaturally hard due to the layer mask.  Normally, the edges of out-of-focus objects appear very soft.  We could have prevented this by feathering (section 10.6) just that part of the selection prior to creating the layer mask, but we can achieve the same effect now by using a soft-edged brush to paint along the edge of the layer mask.  The layer mask is black and white (actually, grayscale), so when softening edges in the layer mask you just need to choose a black or white color for the brush (or a shade of gray).  Click on the layer mask itself (to indicate to Photoshop that you want to modify the pixels of the mask rather than the actual image pixels of the layer) and then simply paint with the soft-edged brush in the main window in Photoshop.  As you paint you’ll see the pattern in the layer mask changing, and you’ll also see the effects of changes to the layer’s transparency, as more or fewer of the pixels from the lower layers show through (in the main Photoshop window).  The figure below shows the result of softening the edges of the distant part of the branch in this way.  The near part of the branch was also repaired using the Clone tool. 


    Note that softening the edges of a layer mask can also be done by selecting parts of the layer mask and applying Gaussian Blur.  Indeed, just about any adjustment that you can make to a layer’s pixels can also be made to a layer’s mask, since the mask is itself nothing more than a grayscale image.  The black pixels in the layer mask cause the corresponding pixels in the layer to be transparent, while the white pixels in the mask cause those pixels in the layer to be opaque; shades of gray impose corresponding degrees of opacity and hence cause blending.  Applying effects like Gaussian Blur to the mask alters the mask’s pixels, which in turn alters the transparency pattern of the layer.  In section 13.2 we’ll use this fact to create artificial clouds in the sky by rendering cloud patterns into the layer mask, so that lighter pixels from a lower level show through those semi-transparent areas.  A similar trick can be used to add texture to a background by duplicating the background, modifying the brightness of the lower layer, and then painting into the upper layer’s mask with an appropriately textured brush.
    The figure below shows a more challenging example, where more subtle blending of the foreground and background layers was required in the wings, due to the motion blur.  This was achieved in this case by painstakingly brushing over appropriate parts of the wings in the background layer mask with a soft, low-opacity brush to allow the background to show through a bit in those regions; it’s likely, however, that a simpler approach could have been found, such as via the use of blurring filters (applied to the layer mask) or feathering (applied earlier, during selection).  Remember that there are usually multiple ways to achieve the same thing in Photoshop, and some of those ways often take far less effort than others.



     If you look closely enough at the above figure, you’ll surely find some processing artifacts (such as the thin, dark line along the belly).  After performing major image surgury like that illustrated here, it’s a good idea to go over the image with a fine-toothed comb to search for any artifacts that need to be cleaned up after the fact.  I recommend zooming in one or two factors of magnification and then panning over the whole image to look for defects.  Some defects you’ll be able to repair fairly simply, by modifying a few pixels in a layer mask.  Others will require the use of the Clone tool.  If you haven’t yet mastered the Clone tool, then it’s a good time to start getting more experience with it; many advanced post-processing techniques depend crucially on either this tool or on the use of related methods that result in copying pixels from one place in the image to another (or from another image altogether, as in the case of replacing a scene’s natural background).
    In this section we’ve relied heavily on the use of layers and layer masks.  It’s worthwhile to just briefly mention two related layer-oriented tricks that you can use in your post-processing, either to simplify things or to reduce memory requirements (or both).  Duplicating a layer obviously incurs a memory cost, since all of the pixels in the original layer now need to be represented twice.  This will result in larger file sizes and in greater demands on run-time memory, which may or may not be an issue for you, depending on how souped-up your computer is.  The first trick is the use of smart objects in Photoshop.  When clicking the Open Image button in Adobe Camera Raw (ACR), if you hold down the Shift key while clicking the button, the Open Image button should change to say Open Object.  The image will then be opened in Photoshop as a smart object rather than as a standard image file.  The difference is that any filters or effects that you apply to the image will be stacked into a set of smart filters to be dynamically evaluated during rendering of the image, rather than being applied statically to the layer’s pixels right then and there.  The figure below shows the Layers panel after three image effects (Shadows/Highlights, Reduce Noise, and Unsharp Mask) were been applied to a smart object.



    The advantage of smart filters is that they automatically update the image whenever anything in the filter chain changes.  For example, if you insert a new filter partway through the filter chain (say, between Unsharp Mask and Reduce Noise in the example above), all filters above this in the chain will be dynamically re-evaluated and the sum effect on the rendered image will appear on screen in “near real time” (though on many computers this can be a slow process, so “near real time” is rather relative).  The importance of ordering in the filter chain should be obvious if you think about a simple example: sharpening versus noise reduction.  Since sharpening emphasizes noise, sharpening before noise reduction can make the noise reduction less effective.  On the other hand, noise reduction obliterates details, so sharpening after noise reduction may result in less subject detail.  Deciding on the ideal ordering may require you to consider both orderings as well as different parameter settings for both filters.  The point is simply that ordering does matter, and with smart filters you can modify that ordering by simply dragging the filters up or down in the stack using your mouse.
    Smart objects and smart filters fall under the rubric of what digital artists call nondestructive editing—that is, editing operations that don’t actually modify the original, underlying representation, so that they can be undone at any point in the future.  By comparison, although the history facility in Photoshop (section 10.5) provides a sort of “undo” capability, undoing something far back in time requires that everything done between then and now also be undone.  Editing in this paradigm is very linear (in time).  The goal of nondestructive editing is to allow adjustments performed arbitrarily far back in time to be re-parameterized as needed.  For example, in the above figure we sharpened the image prior to reducing noise and then adjusting the shadows and highlights.  If we then decide that the image looks a bit over-sharpened, we can go back and adjust the sharpening parameters (e.g., the amount and radius) without having to undo the noise reduction and shadow/highlight adjustment.  The changes to the sharpening parameters will automatically propagate through the smart-filter chain, and the result will be apparent on-screen in near real-time.  I personally don’t use smart objects much, because on my computer (a newest generation Apple laptop) there is still a bit more latency in the updates of the filter chain than I’d like.  But I still think it’s a great idea, and if I wasn’t so lazy I’d spend more time trying to get used to using it.
    Another great idea that I personally don’t use much (but probably should) is the adjustment layer in Photoshop.  Adjustment layers are basically just like smart filters, except that they can be applied to any image, not just to smart objects.  In Photoshop version CS3, smart filters are primarily limited to true filters (under the Filter menu), largely precluding other operations such as those under Image > Adjustments.  Adjustment layers remedy this shortcoming by permitting non-filter image adjustments to be applied in a manner just like smart filters.  The figure below shows an example.  The smart object is the lowest layer; above this we’ve applied two adjustment layers: Levels, and then Hue/Saturation.  The parameters of either of these adjustments can be changed at any time by simply double-clicking the adjustment layer, which will bring up the appropriate adjustment dialog with all the parameter sliders.  You can then change any of the sliders, press Enter/Return and see the result as the image is automatically updated to reflect the parameter changes.  Note also that adjustment layers can have their own layer masks, so they can be applied to specific parts of an image, just as with traditional selection-based processing (i.e., the D-PIE technique described earlier: Differential Processing of Image Elements).


    Smart filters and adjustment layers can be great time savers, because they allow you to go back and correct parameter settings long after you originally applied the effect.  Without these special features, you’d have to undo everything since the original effect was applied, redo the effect with the new parameters, and then replicate all of your processing since them.  One thing that I don’t like about these dynamic filter chains is that the preview that I see on-screen sometimes seems to differ in subtle ways from what I see when I export the image as a JPG and then view the resulting JPG file statically.  Whether this is still an issue with the newest versions of Photoshop is unknown to me at this point.  Also, it’s entirely conceivable that there may be instances in which you wouldn’t want changes to layers lower in the stack to be reflected in higher layers; this would presumably require a change back to the opaque layer method—i.e., avoiding the use of an adjustment layer or smart filter.