Inverse Discrete Cosine Transform Rounding Error Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image and video coding standards face challenges in accurately compressing and decompressing high-resolution frames due to computational difficulties in discrete cosine transforms (DCTs) and inverse DCTs, leading to rounding errors and 'IDCT drift' caused by limited precision numbers and quantization.
Innovation Solution
The techniques involve approximating the computation of inverse discrete cosine transforms using fixed-point arithmetic, employing scaling and biasing methods to reduce rounding errors, and applying butterfly structure operations with rational approximations of irrational constants to maintain precision in smaller devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If real numbers are used in FDCT and IDCT computations, then computation accuracy is improved, but computational complexity and resource requirements increase
Solution Approach 1:
The patent transforms the mathematical parameters of the IDCT computation from real numbers to fixed-point numbers with specific precision (e.g., 16-bit fixed-point). This parameter change allows the computation to be performed using integer arithmetic operations instead of floating-point operations, reducing computational complexity while maintaining acceptable accuracy for video compression applications.
Solution Approach 2:
The patent replaces the mathematical computation mechanism from floating-point arithmetic to fixed-point arithmetic. This substitution changes the underlying computational mechanism to use integer operations, bit shifting, and addition/subtraction instead of multiplication and division, thereby reducing the hardware requirements and computational complexity.
2Device complexity
If limited precision numbers are used to approximate real numbers in IDCT, then computational complexity is reduced, but rounding errors increase
Solution Approach 1:
The patent applies preliminary scaling to the input coefficients before performing the IDCT computation. By pre-scaling the coefficients to appropriate fixed-point representations, the patent prepares the data in advance to minimize rounding errors during subsequent computation steps. This preliminary action ensures that the fixed-point arithmetic maintains sufficient precision throughout the transformation process.
Solution Approach 2:
The patent incorporates bias terms and offset adjustments in the fixed-point IDCT computation to cushion against rounding errors. By adding these compensatory values beforehand in the computation sequence, the patent prevents error accumulation and ensures that the final pixel values remain accurate despite using limited precision arithmetic.
3Productivity
If quantization is applied during compression, then compression ratio is improved, but decomposition accuracy deteriorates
Solution Approach 1:
The patent uses different quantization parameters for encoding and decoding to compensate for quantization errors. By adjusting the quantization step sizes and applying inverse quantization with compensatory factors during decomposition, the patent mitigates the accuracy loss introduced by quantization while maintaining high compression ratios.
Solution Approach 2:
The patent incorporates feedback mechanisms in the IDCT computation where the decoded values are compared against expected ranges and adjusted accordingly. This feedback loop helps correct quantization-induced errors by using the known properties of the transform and the statistics of natural images to guide the reconstruction process.
4Measurement precision
If IDCT drift occurs in predictive frames, then error accumulation increases, but maintaining precision increases computational burden
Solution Approach 1:
The patent applies preliminary corrections to the fixed-point IDCT computation to prevent error accumulation in predictive frames. By incorporating compensation terms and using symmetric rounding modes that prevent systematic bias, the patent counteracts the development of IDCT drift before it can accumulate across multiple frames, thereby maintaining long-term decoding precision without requiring complex error correction mechanisms.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Techniques are described to reduce rounding errors during computation of discrete cosine transform using fixed-point calculations. According to these techniques, an inverse discrete cosine transform a vector of coefficients is calculated using a series of butterfly structure operations on fixed-point numbers. Next, a midpoint bias value and a supplemental bias value are added to a DC coefficient of the matrix of scaled coefficients. Next, an inverse discrete cosine transform is applied to the resulting matrix of scaled coefficients. Values in the resulting matrix are then right-shifted in order to derive a matrix of pixel component values. As described herein, the addition of the supplemental bias value to the DC coefficient reduces rounding errors attributable to this right-shifting. As a result, a final version of a digital media file decompressed using these techniques may more closely resemble an original version of a digital media file.