The automatic feature detection can be driven manually by adding a bias to the pixels of the image.
For example, it is possible to protect regions of the image by adding a positive bias to the corresponding pixels. This will make the seams more unlikely to cross those regions, thus avoiding distortion (but increasing distortion of the other regions).
It is also possible to make the seams more likely to cross some regions by adding a negative bias to them. In this case, reducing the size of the image will tend to erase those regions, while (possibly) keeping the rest of the image in a consistent state.
The bias has to be added always after
the LqrCarver initialization and before
resizing takes place.
In all of the bias-related functions, the bias is added on top of the existing one, so that all of the functions can be called multiple times.
(See also the section "Additional notes" in the README file for situations where the behaviour of biased regions can be counter-intuitive.) [XXX]
The bias values are stored as an array of floating points. It is possible to use directly one such array through this function:
LqrRetVal lqr_carver_bias_add( | LqrCarver* | carver, |
| gdouble* | buffer, | |
| gint | bias_factor); |
Here, buffer is an array contining the bias values,
and it is assumed to have the same size as the image loaded in
carver, while bias_factor is an
overall bias factor, which can be used to affect the global
bias level: if the elements of buffer are of order 1, a standard
choice for the bias_factor would be between 100 and 1000.
This function, and all the following, will not swallow
the buffer (to the contrary of what happens e.g. when creating
a new LqrCarver object).
The bias can also be read from an rgb buffer. This buffer has to be in the same format as the one used in the LqrCarver constructor (but may have a different number of colour per channel). The function is:
LqrRetVal lqr_carver_bias_add_rgb( | LqrCarver* | carver, |
| guchar* | buffer, | |
| gint | bias_factor, | |
| gint | bpp); |
As in the previous case, buffer is assumed to
hold and image of the same size as the one in the carver.
The buffer contents will be transformed into
floating-points by averaging the colour components and multiplying the
result by the alpha channel (transparency) value.
The existence of an alpha channel is inferred from the
bpp value: if this is 1 or 3, no alpha
channel is assumed, if it is 2 or 4, it is assumed that the last
channel is holds the alpha value.
The above functions operate on the whole LqrCarver image.
It is also possible to access specific image regions; for
the floating point use:
LqrRetVal lqr_carver_bias_add_area( | LqrCarver* | carver, |
| gdouble* | buffer, | |
| gint | bias_factor, | |
| gint | width, | |
| gint | height, | |
| gint | x_off, | |
| gint | y_off); |
while for the rgb image use:
LqrRetVal lqr_carver_bias_add_rgb_area( | LqrCarver* | carver, |
| gdouble* | buffer, | |
| gint | bias_factor, | |
| gint | bpp, | |
| gint | width, | |
| gint | height, | |
| gint | x_off, | |
| gint | y_off); |
In both functions, width and height
are used to specify the size of the area of interest, while
x_off and y_off
specify its offset.
For the rest, both functions work in the same way as their global couterpart.
The provided buffers have to be of size
width * heightwidth * height * bppLqrCarver image area : only the parts which overlap with it will be used.
For example, the offsets can also be negative.