Overview
Core Image abstracts the pixel-level manipulation process required when applying a filter to an image, making it simple for applications to implement image transformation capabilities without extensive coding. In a simple implementation, Core Image applies a single Image Filter to the pixel data of a given source to produce the transformed image. Each Image Filter specifies a single transform or effect, either built into Core Image or loaded from a plugin called an Image Unit. Combined with preset or user-defined input parameters, the filter can be applied to the original pixel data without modifying it, thereby providing non-destructive image editing.
Like Photoshop, Core Image can apply multiple filters to the same image source. Instead of applying a series of filters individually, Core Image assembles a dynamic instruction pipeline so that only one calculation needs to be applied to the pixel data to achieve a cumulative effect. Applying the pixel operations associated with multiple filters can be achieved simultaneously and without a significant increase in processing time. Regardless of the number of filters, Core Image assembles the code for this instruction pipeline with a just-in-time compiler, which is executed by either the CPU or graphics card's GPU, whichever can perform the calculation faster.
Filters are written in the Core Image Kernel Language, which shares a subset of commands with OpenGL Shading Language (GLSL). When a compatible GPU is available, the Core Image compiler writes the instruction pipeline using GLSL, handling buffers and states transparently. Although GPU rendering is preferred, the compiler can operate in a CPU fallback mode, generating commands suitable for the current CPU architecture instead. CPU fallback uses the vector processing capabilities of the current CPU or CPUs, and it is multi-processor aware. Thus, Core Image performance depends on the GLSL capabilities of the GPU or the processing power of the CPU. With a supported GPU, most effects can be rendered in realtime or near-realtime.
Read more about this topic: Core Image