Abstract:The Karhunen-Loève transform (KLT) stands as a well-established discrete transform, demonstrating optimal characteristics in data decorrelation and dimensionality reduction. Its ability to condense energy compression into a select few main components has rendered it instrumental in various applications within image compression frameworks. However, computing the KLT depends on the covariance matrix of the input data, which makes it difficult to develop fast algorithms for its implementation. Approximations for the KLT, utilizing specific rounding functions, have been introduced to reduce its computational complexity. Therefore, our paper introduces a category of low-complexity, data-independent KLT approximations, employing a range of round-off functions. The design methodology of the approximate transform is defined for any block-length $N$, but emphasis is given to transforms of $N = 8$ due to its wide use in image and video compression. The proposed transforms perform well when compared to the exact KLT and approximations considering classical performance measures. For particular scenarios, our proposed transforms demonstrated superior performance when compared to KLT approximations documented in the literature. We also developed fast algorithms for the proposed transforms, further reducing the arithmetic cost associated with their implementation. Evaluation of field programmable gate array (FPGA) hardware implementation metrics was conducted. Practical applications in image encoding showed the relevance of the proposed transforms. In fact, we showed that one of the proposed transforms outperformed the exact KLT given certain compression ratios.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the computational complexity and data - dependence issues of the Karhunen - Loève Transform (KLT) in practical applications. Specifically: 1. **Computational Complexity**: The calculation of KLT depends on the covariance matrix of the input data, which makes its computational complexity high and it is difficult to design an efficient fast algorithm. 2. **Data Dependence**: Since the transformation matrix of KLT depends on the input data, different input signals will lead to different transformation matrices, so the transformation basis required by the decoder cannot be determined in advance, that is, the so - called dictionary exchange problem. To solve these problems, the paper proposes a class of KLT approximation methods with low complexity and independent of input data, which are designed based on a series of rounding functions. Through this method, the author aims to reduce the computational complexity of KLT and overcome its data - dependence, making it more suitable for practical applications, especially in the fields of image and video compression. ### Main Objectives The main objective of the paper is to propose low - complexity KLT approximate transformations applicable to different correlation coefficients $\rho$ so as to be able to handle medium - and low - correlated signals. Specifically, the author hopes to achieve this goal through the following characteristics: - **Data Independence and Closed - form Expression**: The transformation does not depend on the input data and has a clear mathematical expression. - **Symmetric Structure**: This structure can lead to sparse matrix decomposition and fast algorithms. - **Applicable to First - order Markov Processes**: It performs well in a wide range of correlation coefficients. ### Methodological Innovation The methodology proposed by the author is unprecedented in the literature, especially for the application of KLT in first - order Markov processes. Through this method, the author hopes to provide a new, low - complexity KLT approximation method adaptable to multiple contexts. ### Experimental Verification The paper also experimentally verifies the performance of the proposed approximate transformations in image compression, and the results show that these approximate transformations are even better than the exact KLT and DCT in some cases. In summary, the core problem of this paper is to solve the computational complexity and data - dependence problems of KLT in practical applications by introducing low - complexity, data - independent KLT approximation methods, thereby improving its applicability in fields such as image and video compression.

Fast Data-independent KLT Approximations Based on Integer Functions