Block adaptive linear predictive coding with adaptive gain and bias

Claim 2

An improved block adaptive predictive coding method of the type wherein a quantizer is selected from a finite set of quantizers to encode a block of error signals based on the statistics of the error signals in the block, wherein the improvement comprises: selecting block quantizers based on a plurality of statistical parameters.

Claim 3

The method claimed in claim 1, wherein the statistical parameters are a statistical estimate of the range of error signals in a block and a statistical estimate of the centroid of the error signal in the block.

Claim 4

The method claimed in claim 2, wherein the estimate of the range is the standard deviation (σ) and the estimate of the centroid is the mean (μ).

Claim 5

The method claimed in claim 2, wherein the estimate of the range is F(max-min) and the estimate of the centroid is the mode.

Claim 6

An improved block adaptive linear predictive coding method for compressing and transmitting or storing a digital image of the type including the steps of: dividing the image into 1×B blocks of pixels; forming a two-dimensional prediction for each pixel value based upon a linear combination of neighboring pixel values; subtracting successive predictions from actual pixel values to form a one-dimensional block of error signals; calculating the sigma (σ) of the error signals in each block; selecting a quantizer for each block from a set of scaled sigma Max quantizers using the calculated sigma; quantizing the error signals in each block with the selected quantizer; transmitting or storing the first line of an image and the first pixel in each line, a code word for the quantizer, and the quantized error signals as the compressed image, wherein the improvement comprises the steps of: (a) calculating the mean (μ) of the error signals in each block; and (b) selecting the quantizer for each block using the calculated sigma and the calculated mean from a set of minimum square error two-variable quantizers based on probability models for a function of the mean (μ) and sigma (σ) values.

Claim 7

An improved block adaptive linear predictive coding method for compressing and transmitting or storing a digital image of the type including the steps of: dividing the image into 1×B blocks of pixels; forming a two-dimensional prediction for each pixel value based upon a linear combination of neighboring pixel values; subtracting successive predictions from actual pixel values to form a one-dimensional block of error signals; calculating the sigma (σ) of the error signals in each block; selecting a quantizer for each block from a set of scaled sigma Max quantizers using the calculated sigma; quantizing the error signals in each block with the selected quantizer; transmitting or storing the first line of an image and the first pixel in each line, a code word for the quantizer, and the quantized error signals as the compressed image, wherein the improvement comprises the steps of: (a) calculating the mean (μ) of the error signals in each block; and (b) selecting the quantizer for each block using the calculated sigma and the calculated mean from a set of minimum square error two-variable quantizers based on probability models for a function of the mean (μ) and sigma (σ) values.

Claim 8

The method claimed in claim 5, wherein the mean (μ) and signal (σ) values are employed to select the quantizers, and the probability model for the mean (μ) is a Laplacian, and the probability model for sigma (σ) is an exponential.

Claim 9

The method claimed in claim 5, further comprising the steps of: (a) performing a polar transformation on the mean (μ) and sigma (σ) values, where the polar angle is arctan σ/μ and the polar radius is √μ 2 +σ 2 ; and (b) The transformed polar angle and polar radius values are employed to select a set of minimum square error two-variable quantizers based upon a uniform probability model for the polar angle and an exponential probability model for the polar radius.

Claim 10

The method claimed in claim 5, 6, or 7, wherein the quantizers have different numbers of quantization levels based on the value of σ, higher values of σ having more quantization levels.

Claim 11

The method claimed in claim 8, further including the step of Huffman encoding the code word for the quantizer and the error signals.

Claim 12

An improved block adaptive linear predictive coding method for compressing and transmitting or storing a digital image of the type including the steps of: dividing the image into 1×B blocks of pixels; forming a two-dimensional prediction for each pixel value based upon a linear combination of neighboring pixel values; subtracting successive predictions from actual pixel values for form a one-dimensional block of error signals; calculating the sigma (σ) of the error signals in each block; selecting a quantizer for each block from a set of scaled sigma Max quantizers using the calculated sigma; quantizing the error signals in each block with the selected quantizer; transmitting or storing the first line of an image and the first pixel in each line, a code word for the quantizer and the quantized error signals as the compressed image, wherein the improvement comprises the steps of: (a) prior to the step of calculating the sigma value, comparing the average absolute value S of the difference between the input signal values and the average A of their three nearest reconstructed neighbors in a block with the threshold value T, and if S is less than T, treating the block as one in a run of blocks for which no sigma value is calculated, no quantizer is selected and no error signal is transmitted; and (b) generating and transmitting a flag indicating a run of blocks and a value indicating the length of the run.

Claim 13

An improved block adaptive linear predictive coding method for compressing and transmitting or storing a digital image of the type including the steps of: dividing the image into 1×B blocks of pixels; forming a two-dimensional prediction for each pixel value based upon a linear combination of neighboring pixel values; subtracting successive predictions from actual pixel values for form a one-dimensional block of error signals; calculating the sigma (σ) of the error signals in each block; selecting a quantizer for each block from a set of scaled sigma Max quantizers using the calculated sigma; quantizing the error signals in each block with the selected quantizer; transmitting or storing the first line of an image and the first pixel in each line, a code word for the quantizer and the quantized error signals as the compressed image, wherein the improvement comprises the steps of: (a) prior to the step of calculating the sigma value, comparing the average absolute value S of the difference between the input signal values and the average A of their three nearest reconstructed neighbors in a block with the threshold value T, and if S is less than T, treating the block as one in a run of blocks for which no sigma value is calculated, no quantizer is selected and no error signal is transmitted; and (b) generating and transmitting a flag indicating a run of blocks and a value indicating the length of the run.

Claim 14

The method claimed in claim 10, further including the steps of: (a) calculating the mean (μ) of the error signals in each block; and (b) selecting the quantizer for each block using the calculated sigma and the calculated mean from a set of minimum square error two-variable quantizers based on probability models for a function of the mean (μ) and sigma (σ) values.

Claim 15

The method claimed in claim 11, wherein the mean (μ) and sigma (σ) values are employed to select the quantizers, and the probability model for the mean (μ) is a Laplacian and the probability model for sigma (σ) is an exponential.

Claim 16

The method claimed in claim 11, further including the steps of: (a) performing a polar transformation on the mean (μ) and sigma (σ) values, where the polar angle is arctan σ/μ and the polar radius is √μ 2 +σ 2 ; and (b) The transformed polar angle and polar radius values are employed to select from a set of minimum square error two-variable quantizers based upon a uniform probability model for the polar angle and an exponential probability model for the polar radius.

Claim 19

The method claimed in claim 15, further including the steps of Huffman encoding the code word for the quantizer, the error signals, and the run lengths.

Claim 20

An improved block adaptive linear predictive coding method for compressing and transmitting or storing a digital image of the type including the steps of: dividing the image into 1×B blocks of pixels; forming a two-dimensional prediction for each pixel value based upon a linear combination of neighboring pixel values; subtracting successive predictions from actual pixel values for form a one-dimensional block of error signals; calculating the sigma (σ) of the error signals in each block; selecting a quantizer for each block from a set of scaled sigma Max quantizers using the calculated sigma; quantizing the error signals in each block with the selected quantizer; transmitting or storing the first line of an image and the first pixel in each line, a code word for the quantizer and the quantized error signals as the compressed image, wherein the improvement comprises: The quantizers having different numbers of quantization levels based on the value of σ, higher values of σ having more quantization levels.

Claim 21

An improved block adaptive linear predictive coding method for compressing and transmitting or storing a digital image of the type including the steps of: dividing the image into 1×B blocks of pixels; forming a two-dimensional prediction for each pixel value based upon a linear combination of neighboring pixel values; subtracting successive predictions from actual pixel values for form a one-dimensional block of error signals; calculating the sigma (σ) of the error signals in each block; selecting a quantizer for each block from a set of scaled sigma Max quantizers using the calculated sigma; quantizing the error signals in each block with the selected quantizer; transmitting or storing the first line of an image and the first pixel in each line, a code word for the quantizer and the quantized error signals as the compressed image, wherein the improvement comprises: The quantizers having different numbers of quantization levels based on the value of σ, higher values of σ having more quantization levels.