A parallel vector quantization processor eliminating redundant calculations for real-time motion picture compression

A generalized multiple projection axes method for fast encoding of vector quantization

“…Image block is 4 · 4 or k = 16. Codebooks of size 256, 512 and 1024 are generated using Lena image as a training set based on the modified KohonenÕs self-organizing map (SOM) method developed in (Nozawa et al, 2000). Suppose the elements in an image block are arranged in a raster order.…”

A 66fps 3 8mW nearest neighbor matching processor with hierarchical VQ algorithm for real-time object recognition

“…In order to decrease the search region and processing time, some of previous VQ processors employ hierarchical approach like two-step search or tree search [8], [10]. However, both of two VQ algorithms have the same problem of accuracy loss.…”

“…However, these VQ processors suffer from large silicon area caused by expanded processing logic elements for distance calculation. And one of them [8] generates computational error in VQ operation instead of obtaining computational benefits. However, different from previous VQ processors for image compression, the nearest neighbor matching processor cannot utilize full area of the chip and dedicated off-chip memories because it is a sub system block of object recognition SoC.…”

Fast Encoding Method for Vector Quantization Based on a New Mixed Pyramid Data Structure

“…Codebooks of size 256, 512 and 1024 are generated using 512 512, 8-bit Lena image as a training set based on [10]. Block size is 4 4.…”

“…Let P=(1,1, … ,1) in R k space be a projection axis in R k space, which is called as a central axis in [3]. From a viewpoint of the geometry, sum is the projection component of a vector onto P and variance is the orthogonal component from this vector to sum [10]. In other words, sum and variance of a vector can be viewed as an orthogonal decomposition for a vector on the axis P. Because sum and variance are orthogonal in R k space, it is beneficial to combine them into the pyramid data structure.…”