China
Africa
Explore chapters and articles related to this topic
Multimedia Data Compression
Published in Sreeparna Banerjee, Elements of Multimedia, 2019
Transform coding converts data into a form where compression is simpler to implement. This transformation changes the symbols, which are correlated into a representation where they are decorrelated. The new values are usually smaller on average than the original values. The net effect is to reduce the redundancy of the representation. The transform coefficients can then be quantized according to their statistical properties, producing a much-compressed representation of the original data. Because this coding technique eliminates data, it is a lossy process. Some popular transform coding techniques are DCT and discrete wavelet transform, which will be discussed in the following section.
Image Compression
Published in Scott E. Umbaugh, Digital Image Processing and Analysis, 2017
Transform coding is a form of block coding done in the transform domain. The image is divided into blocks, or subimages, and the transform is calculated for each block. Any of the previously defined transforms can be used, frequency or sequency, but during the development of the original JPEG algorithm, it was determined that the discrete cosine transform (DCT) is optimal for most images. The next extension of the JPEG algorithm, JPEG2000, uses the wavelet transform, which has been found to provide even better compression, and wavelet compression is explored in the next section.
Part Overview: Coding of Video and Multimedia Content
Published in Ling Guan, Yifeng He, Sun-Yuan Kung, Multimedia Image and Video Processing, 2012
Quantization plays a key role in any lossy coding scheme: it offers a highly reduced bit rate on the one hand (because many coefficients become zero after quantization) but is also perhaps the sole source to the resulted coding distortion, on the other. So far, quantization has been kept rather simple such that a uniform quantizer (with or without a dead zone) is commonly employed in various scenarios, especially in the block-based transform coding schemes; whereas the optimal (nonuniform) quantizers can be designed once some necessary statistical knowledge of the source is known.
An image compression approach for efficient pneumonia recognition
Published in The Imaging Science Journal, 2023
Sabrina Nefoussi, Abdenour Amamra, Idir Amine Amarouche
The motivation behind transform coding is that simple coding can be more efficient in the transform domain than in the original space [23]. In light of this principle, we aim to extend image coding efficiency. Besides the transmission and storage, encoding and compressing the contents of a 2-D image could contribute to classification and decision support. A compressed image is represented in bits, with the natural aim to keep the number of bits used as low as possible while preserving the regularity of latent space. In principle, these bits carry the essentials that best describe the image features, thus allowing the recovery of any useful information while ignoring irrelevant ones. The contribution of autoencoders in the area of image denoising strengthens our idea, as they can be deemed as the best pre-processing technique, based on deep neural networks, for image classification [24]. The whole idea of our work lies in that if we can reconstruct an entire image from a reduced amount of information (compressed image), therefore necessarily, this same amount encompasses the discriminating characteristics of an image that belongs to a class A, which can characterize it from another image that belongs to another class B. But now, the real challenge is how to proceed to accentuate these discriminating features in a code (of encoded image) to guide the classification; this is where the difficulty lies.
Compression of Assembly Power Form Factors for Core Calculations
Published in Nuclear Science and Engineering, 2019
In this work, two transform methods are applied on the 2-D maps of the power form factors, which are considered here as raw and raster images with the power values as color intensity or brightness of each pin cell in the lattice. The main idea behind transform coding is that the transformed coefficients are less correlated and the information is compacted into a small proper subset of transform coefficients.17 The original power values are thence decomposed on a weighted sum of basic discrete functions, where the corresponding weights are the transform coefficients. These represent a measure of the correlation between the original and the basic distributions.