MimConvolve

This function performs a general convolution operation on the source buffer using the specified filter, storing results in the specified destination buffer. This function supports two types of filters: Finite Impulse Response (FIR) filters and Infinite Impulse Response (IIR) filters. FIR filters operate on a finite neighborhood, while IIR filters take in account all values in an image. You can specify either a predefined or a custom FIR or IIR filter.

This function provides a number of predefined FIR. For predefined FIR filters, the overscan is set to transparent overscan pixel values, saturation is enabled for results, and the kernel's default center pixel is set as the top-left pixel of the central elements in a neighborhood. When using predefined FIR filters, you cannot change operation control settings except for the overscan.

This function provides two types of predefined IIR filters. Predefined IIR filters, M_SHEN_FILTER() and M_DERICHE_FILTER(), use a recursive implementation, and the overscan pixel values are ignored. The normalization factor is automatically set so that it is possible to use the full data range of the destination buffer without overflows, and the result is given in the number of bits of the destination buffer. When using predefined IIR filters, you cannot change operation control settings except for the smoothness value. To use an IIR filter in kernel mode, you must use a custom IIR filter.

This function also accepts a custom FIR or IIR filter with which you can customize operation control settings. To define a custom filter, you must allocate a kernel buffer using MbufAlloc1d() or MbufAlloc2d(). Note that the kernel buffer size can be constrained by the available resources.

To define a custom FIR filter, you must load the kernel buffer with values, using MbufPut(). To define a custom IIR filter, use MbufControlNeighborhood() with M_FILTER_TYPE, M_FILTER_OPERATION, and M_FILTER_MODE; the kernel values are automatically calculated based on these settings.

When using custom FIR or IIR filters, you can change default operation control settings using MbufControlNeighborhood().

For predefined and custom FIR filters, when using 32-bit integer buffers and saturation is disabled, the accumulator buffer is a 32-bit integer buffer (MIL_INT); whereas, when saturation is enabled, the accumulator buffer is a 64-bit floating-point internal buffer (double). For IIR filters, when using 32-bit floating-point source or destination images, operations are performed using floating-point precision; whereas, for other types of images, operations are performed using fixed-point precision.

For custom FIR filters, this function will internally separate a large kernel into two 1-dimensional kernels, if possible, to increase the speed of the convolution operation. For more information on separating kernels, see the Finite Impulse Response (FIR) filters subsection of the Custom spatial filters section of Chapter 4: Advanced image processing.

Custom FIR filters containing a unique value (such as uniform kernels or average kernels) are dealt with by a specifically targeted algorithmic optimization.

If the source and destination are multi-band buffers, the same filter is applied to every band of the source buffer.