190 likes | 325 Views
This resource provides comprehensive insights into handling and visualizing multispectral and hyperspectral data. Users can download datasets and learn how to display spectral signatures using example MATLAB code. It delves into scalar quantization concepts, including encoder and decoder mapping. The material explains various quantizers, emphasizing uniform and non-uniform techniques, along with the significance of mean-square quantization error. Additionally, applications like prediction-based and transform-based coding in spectral data processing are discussed, offering practical insights for remote sensing analysis. ###
E N D
Read spectral data • From www.ee.unlv.edu/~regent/SPACE/DATA • Download multispectral data: mutispectral_data.mat • Download hyperspectral spectral data: hyperdata.mat • Display bands: Example: I=uint8(data(:,:,1)); %read first band figure, imshow(I); Assignment: 1.Display spectral signature of the pixel at (30, 40)
Scalar Quantization Dr.E.Regentova
Quantization • To represent a large set of values with a much smaller set • Encoder mapping • The encoder divides the range of source into a number of intervals. • Each interval is represented by a distinct codeword. • Decoder mapping • For each received codeword the decoder generates a reconstruct value. • The set of output can be scalars or vectors • Type of quantizers: scalar and vector
Definition of Scalar Quantization • M-level quantizer is typically characterized by M+1 decision levels, or boundaries, b0,b1, ... , bM, and by M reconstruction levels y0,y1, ... yM-1. • The bi’s divide the range of data under quantization into k consecutive intervals [b0,b1) [b1,b2) ... [bM-1 bM). • Each yi is in [bi,bi+1), and can be viewed as the ”centroid” of its interval.
Quantizing a number yi means locating the interval [bi,bi+1) that contains yiand replacing it by index i. • Dequantization (in reconstruction) is the process of replacing each index i by the value ri. This approximates every original number that was in interval [bi,bi+1) by the centroid ri.
MSQE - mean-square quantization error If the quantization operation is Q If the input is modeled by a random variable X with pdf f(x)
Code rate • The selection of number of intervals M depends on R to be achieved. • For the fixed-length code, decision and reconstruction levels are obtained by minimizing MSE for the given R.
Types of Scalar Quantizers • Uniform Quantizers : All intervals are the same size, except for 2 outer intervals • All the decision intervals are of equal size= (bM-b0)/M • bi=b0+i ▪ • The reconstruction levels ri are the centers of the intervals • ri=(bi+bi+1)/2
Midrise(left) and midthread(right) quantizers M=8 M=7 • - Midrize quantizer • Zero is not one of the output levels • M-is even • Midtread quantizer • zero is one of the output levels • M is odd
Non-uniform Qunatizers • Either the decision intervals are not of equal size • Or the reconstruction levels are not the centers of their intervals Midrise: M=8
Quantization Error: q=x-Q(x) uniform error
Quantization Error: q=x-Q(x) nonuniform error
Application • Prediction based coding: for coding the prediction error • Uusually non-uniform quantizer because of the distribution (modeled as Gaussian) • Transform based coding (DCT and DWT) • JPEG • JPEG2000
Assignment • PCM-Pulse Code Modulation • positive prediction errors are encoded as 1; negative as -1. • Thus only one bit is used • Implement PCM on a pout.tif image.