In this paper, a non-adaptive optimal transform (NAOT) coding system is proposed. Note that the energy-invariant property in an orthogonal transformation and that the mean squared error (MSE) of a reconstructed image is proportional to the total energy of transform coefficients discarded in the coding process. The NAOT coding system is developed and proved optimal in the sense of minimum average energy loss. Basically, the proposed coding system consists of the following steps. First, obtain the average energy image block from transform image blocks. Second, sort the average energy image block in the descending order by energy where the sorted indices are recorded. Third, specify the number of coefficients, M, to be retained in the coding process. Fourth, the first M sorted indices form a set denoted as SM through which the problem of optimal feature selection in transform coding is solved. Fifth, find a fixed mask AM from set SM which is then used to select M significant transform coefficients in image blocks. Finally, the M selected coefficients are quantized and coded by the order as in SM. To verify the NAOT coding system, simulations are performed on several examples. In the simulation, the optimality and the optimal feature selection in the NAOT coding system are justified. Also, the effectiveness of the proposed SM-based selection approach is compared with the zigzag scan used in the JPEG. For fair comparison, the JPEG is modified to code only M transform coefficients. Simulation results indicate that the performance of SM-based selection approach is superior or identical to the zigzag scan in terms of PSNR. Finally, the performance comparison between the NAOT coding system and the JPEG is made. It suggests that the proposed NAOT coding system is able to trade very little PSNR for significant bit rate reduction when compared with the JPEG. Or it can be said that the JPEG wastes much bit rate to improve very little PSNR on the reconstructed image, when compared with the NAOT coding system.

In this paper, a novel type of neural networks called grey neural network (GNN) is proposed and applied to improve short term load forecasting (STLF) performance. This work is motivated by the following observations: First, the forecasting performance of neural network is affected by the randomness in STLF data. That is, poor performance results from large randomness and vice versa. Second, the grey first-order accumulated generating operation (1-AGO) is reported having randomness reduction property. By the observations, the GNN is proposed and expected to have better STLF performance. The GNN consists of grey 1-AGO, the piecewise linear neural network (PLNN), and grey first-order inverse accumulated generating operation (1-IAGO). Given a set of STLF data, the data is first converted by grey 1-AGO and then is put into the PLNN to perform forecasting. Finally, the predicted load of GNN is obtained through grey 1-IAGO. For comparison, the original STLF data is also put into the PLNN itself. With identical training conditions, the simulation results indicate that with various network structures the GNN, as expected, outperforms the PLNN itself in terms of mean squared error.

In this paper, a grey filtering approach based on GM(1,1) model is proposed. Then the grey filtering is applied to speech enhancement. The fundamental idea in the proposed grey filtering is to relate estimation error of GM(1,1) model to additive noise. The simulation results indicate that the additive noise can be estimated accurately by the proposed grey filtering approach with an appropriate scaling factor. Note that the spectral subtraction approach to speech enhancement is heavily dependent on the accuracy of statistics of additive noise and that the grey filtering is able to estimate additive noise appropriately. A magnitude spectral subtraction (MSS) approach for speech enhancement is proposed where the mechanism to determine the non-speech and speech portions is not required. Two examples are provided to justify the proposed MSS approach based on grey filtering. The simulation results show that the objective of speech enhancement has been achieved by the proposed MSS approach. Besides, the proposed MSS approach is compared with HFR-based approach in [4] and ZP approach in [5]. Simulation results indicate that in most of cases HFR-based and ZP approaches outperform the proposed MSS approach in SNRimp. However, the proposed MSS approach has better subjective listening quality than HFR-based and ZP approaches.