As the development of computational hardware capable of variable precision progresses, the customary requirement of high-precision arithmetic is being relaxed, particularly in automated target recognition (ATR) applic...
详细信息
ISBN:
(纸本)0819429112
As the development of computational hardware capable of variable precision progresses, the customary requirement of high-precision arithmetic is being relaxed, particularly in automated target recognition (ATR) applications. Such algorithms typically admit noisy imagery that often has, for example, two to three bits of noise per eight-bit pixel. Since the resulting precision p of five or six bits (SNR = 1:64 to 1:32) is nonincreasing throughout the course of a computational cascade, it is reasonable to assume that p = 8 bits could suffice for most ATR applications. Practical design constraints in addition to noise include limited processor size, weight, and power supply, as well as available computational bandwidth and frame rate. Although limited precision computation can decrease size and power requirement as well as computational cost, the accrual of computational error can severely compromise resultant accuracy, leading to a design tradeoff between error, computational precision, and speed/power requirements that we call the Limited Precision Problem (LPP). In this paper, a restricted instance of the LPP is analyzed namely, the effect of reduced precision on imagecompression transforms. Particular emphasis is placed upon the compounding of representational error in the compression process as computation passes through various stages of a given algorithm. Analysis emphasizes effects of noise and computational error on common compression transforms such as visual pattern imagecoding.(VPIC), vector quantization (VQ), and a recently-developed algorithm called EBLAST [1]. Tests for preservation of input statistics and minimization of mean-squared error (MSE) indicate that, in eight-bit imagery with two to 2.5 bits of noise, as few as five bits of precision suffice for the aforementioned compression algorithms to retain acceptable visual appearance and MSE for ATR operations.
imagecompression is increasingly employed in the exploitation of limited communication channel bandwidth and maximization of storage system efficiency. Although traditional measures of signal or image quality (e.g., ...
详细信息
ISBN:
(纸本)0819429112
imagecompression is increasingly employed in the exploitation of limited communication channel bandwidth and maximization of storage system efficiency. Although traditional measures of signal or image quality (e.g., mean-squared error or MSE) are useful in a communication context, few such measures are evident in the open literature that effectively address implementational issues such as visual quality or suitability for automated target recognition (ATR) of a decompressed image. For example, reconstruction errors such as blocking effect, aliasing or suppression of high frequencies, preservation of image or neighborhood statistics, and quantization-induced perturbation of line segments or area can degrade visual image quality. In numerous ATR applications, such error is an important source of ATR algorithm performance degradation. Unfortunately, the majority of published image quality measures (IQMs) tend to be based on ad hoc or first-order models of human visual system (HVS) function. Such models do not necessarily correlate with the appearance of image details (e.g., higher-order models) or mathematical descriptions of ATR filters (e.g., non-HVS models). Additionally, published IQMs tend not to address the primary ATR problem of feature visibility or the secondary problem of estimating image or neighborhood error distributions in decompressed imagery. In this paper, a collection of performance measures and IQMs designed specifically for imagecompression is presented. Beginning with the customary performance measures of MSE and space-time bandwidth product, we progress to spatial measures such as cutoff frequency, effect of greyscale quantization on variance, measures of texture preservation, and statistics of the modulation transfer function (MTF). Texture is estimated by an area-based variance descriptor derived from the Lorentzian distribution, which has been successfully employed in ATR practice [1]. Measures for disruption of linear features such as line s
This dissertation consists of two parts. The first part presents an independence technique for any m-dependent random variable and its application to datacompression. The second part deals with new types of two-dimen...
详细信息
This dissertation consists of two parts. The first part presents an independence technique for any m-dependent random variable and its application to datacompression. The second part deals with new types of two-dimensional pseudo-noises, quasi m-arrays and the Gold code arrays. These pseudo-noises are used during the application of independence technique to ensure signal encryption. In the first part of this dissertation, a technique to generate independent random variables from the m-dependent ones is proposed and its validity is proved. The probability density function of any resulting independent random variable is also shown to be Gaussian. This technique can be applied to any m-dependent signal, and the resulting independent signal is unrecognizable. Therefore, signal encryption is also achieved. Since these encrypted signals are independent Gaussian random variables, they can be quantized individually. A simple scalar quantizer can be used for these independent Gaussian random variables and the sum of mean-square quantization errors can be minimized. In addition, vector quantization is no longer necessary because these encrypted signals are already independent. Secondly, two-dimensional quasi m-arrays and Gold code arrays are developed. These arrays are easily generated by the modulo-2 addition of m-sequences. The cyclic correlation properties of these arrays are studied. The cyclic auto-correlation of any array is similar to a delta-function. In addition, the cyclic cross-correlation between any two arrays is small compared with their cyclic auto-correlation peaks. Therefore, these arrays have the quasi-orthogonal property. In addition, these arrays can be generated easily which is helpful for many applications. Application of the independence technique to cosine and Hadamard transform imagecoding.is also studied. The quasi m-arrays are used before transform coding.to ensure the signal encryption. Therefore, this technique is named as transform encryption c
暂无评论