Method And System For Lossless Wavelet Decomposition, Compression And Decompression Of Data and Reference File Download Link
https://eu2.contabostorage.com/00f3241116844f24b628f46d81abb929:st1/folder6/6111/1655840489___disclosure_list.xls
2026-05-30 02:12:05 - Admin
<style> body { font-family: Arial, Helvetica, sans-serif; line-height: 1.6; margin: 0; padding: 0 20px; background-color: #f9f9f9; color: #333; } header, main, section { max-width: 800px; margin: auto; padding: 20px 0; } h1, h2, h3 { color: #2c3e50; } pre { background: #eef; padding: 10px; overflow-x: auto; } code { background: #e8e8e8; padding: 2px 4px; border-radius: 3px; } a { color: #2980b9; text-decoration: none; } a:hover { text-decoration: underline; } </style><header> <h1>Method and System for Lossless Wavelet Decomposition, Compression and Decompression</h1></header><main><section> <h2>1. Introduction</h2> <p> Wavelet transforms have become a cornerstone of modern data handling because they provide a hierarchical representation that captures both frequency and location information. When the transform is applied in a reversible (lossless) way, the original data can be perfectly reconstructed after compression, making waveletbased methods suitable for scientific, medical, and archival applications where fidelity cannot be compromised. </p></section><section> <h2>2. Core Concepts</h2> <h3>2.1. Wavelet Decomposition</h3> <p> A onedimensional signal <code>x[n]</code> is processed through a pair of filter banks: a lowpass (scaling) filter <code>h[n]</code> and a highpass (wavelet) filter <code>g[n]</code>. After convolution, the results are downsampled by two, producing the approximation coefficients <code>a[k]</code> and detail coefficients <code>d[k]</code>. Repeating the same steps on the approximation part yields a multilevel decomposition: </p> <pre>a a + da a + daL aL + dL </pre> <p> For lossless operation, the filter coefficients must be rational numbers that can be represented exactly (e.g., integers or fractions) and the lifting scheme is often employed because it replaces convolution by simple addsubtract operations. </p> <h3>2.2. Lifting Scheme</h3> <p> The lifting scheme splits the transform into three steps: <ol> <li><strong>Split:</strong> separate even and odd samples.</li> <li><strong>Predict:</strong> use even samples to predict odd ones; the prediction error becomes a detail coefficient.</li> <li><strong>Update:</strong> modify even samples using the detail coefficients to preserve the average (or other moments).</li> </ol> Because each step is reversible, the inverse transform simply undoes the update, predict, and split steps in reverse order, guaranteeing lossless reconstruction. </p> <h3>2.3. Entropy Coding</h3> <p> After decomposition, many coefficients become zero or nearzero, especially in higher levels. Entropy coders (e.g., Huffman, arithmetic coding, or range coding) exploit this redundancy. For truly lossless compression, the coder must be able to represent every possible integer value without truncation. </p></section><section> <h2>3. System Architecture</h2> <p> A practical implementation follows a pipeline architecture, comprising four logical blocks: </p> <ol> <li><strong>Input Module:</strong> reads raw binary data, partitions it into blocks (e.g., 64KB), and optionally pads the last block.</li> <li><strong>Forward Wavelet Engine:</strong> applies a chosen lossless wavelet (e.g., Integer Haar, Le Gall5/3, or CDF 9/7 with integer lifting) to each block, producing a tree of approximation and detail subbands.</li> <li><strong>Quantisation & Reordering:</strong> for lossless operation, quantisation is omitted. Coefficients are reordered (e.g., by subband) to improve locality for entropy coding.</li> <li><strong>Entropy Coder:</strong> encodes the integer stream using a contextmodelled arithmetic coder. The coder emits a selfcontained bitstream that includes a header describing the wavelet, decomposition depth, and block size.</li> <li><strong>Output Module:</strong> writes the compressed bitstream to storage or transmits it over a channel.</li> </ol> <p> The decoder mirrors the encoder in reverse order, guaranteeing exact reconstruction. </p></section><section> <h2>4. Algorithmic Details</h2> <h3>4.1. Integer Haar Wavelet (example)</h3> <pre>for i = 0 to N/21 a[i] = (x[2*i] + x[2*i+1]) >> 1 // approximation (integer division) d[i] = x[2*i] - x[2*i+1] // detailend for </pre> <p> The inverse simply recovers: </p> <pre>for i = 0 to N/21 x[2*i] = a[i] + ((d[i]+1) >> 1) x[2*i+1] = a[i] - (d[i] >> 1)end for </pre> <h3>4.2. Lifting Implementation for LeGall5/3</h3> <pre>// predict stepfor i = 1 to N2 step 2 d[i/2] = x[i] - ((x[i1] + x[i+1]) >> 1)end for// update stepfor i = 2 to N2 step 2 a[i/2] = x[i] + ((d[(i1)/2] + d[i/2] + 2) >> 2)end for </pre> <p> The integer shifts guarantee that every intermediate value remains an integer, preserving losslessness. </p> <h3>4.3. Adaptive Context Modelling</h3> <p> The coder builds a small context from neighboring coefficients (e.g., previous detail of the same subband) and selects a probability model accordingly. The model is updated after each symbol, allowing the coder to adapt to local statistics in each block. </p></section><section> <h2>5. Practical Considerations</h2> <ul> <li><strong>Block Size:</strong> Larger blocks improve compression ratio but increase memory usage and latency. Typical values range from 32KB to 256KB.</li> <li><strong>Decomposition Levels:</strong> The depth determines how many times the approximation subband is further split. For 2D images, three to five levels are common; for 1D signals, the level is limited by the block length.</li> <li><strong>Parallelism:</strong> Each block can be processed independently, enabling multicore or GPU acceleration. The lifting steps themselves are also amenable to SIMD vectorisation.</li> <li><strong>Metadata:</strong> The bitstream header records: <ul> <li>Wavelet type and lifting coefficients.</li> <li>Number of decomposition levels.</li> <li>Block dimensions.</li> <li>Checksum or CRC for integrity verification.</li> </ul> </li> <li><strong>Error Detection:</strong> Because the method is lossless, any mismatch between original and reconstructed data indicates corruption. A simple bytewise comparison after decoding validates the process.</li> </ul></section><section> <h2>6. Applications</h2> <p> The described method is suitable for any domain where exact reproducibility matters: </p> <ul> <li>Medical imaging (e.g., lossless DICOM storage).</li> <li>Remote sensing and satellite telemetry.</li> <li>Scientific instrumentation (seismic, radar, spectroscopy).</li> <li>Archival of legal documents and highresolution graphics.</li> <li>Hybrid schemes where a lossless core is combined with a later lossy layer for streaming.</li> </ul></section><section> <h2>7. Conclusion</h2> <p> Lossless wavelet decomposition, when implemented with integer lifting schemes and efficient entropy coding, provides a robust framework for compressing a wide variety of data types while guaranteeing exact reconstruction. The modular system described hereinput handling, forward transform, optional reordering, contextadapted entropy coding, and a matching decoderoffers scalability, parallel execution and can be tailored to the constraints of embedded devices or highperformance servers alike. </p> <p> By selecting appropriate wavelets, adjusting block size and decomposition depth, and employing modern adaptive coders, practitioners can achieve compression ratios comparable to conventional lossless algorithms such as PNG or FLAC, while retaining the multiresolution benefits that wavelets uniquely provide. </p></section></main>