Indexes used for evaluation of image quality (EIQ) are provided with computational models to measure the quality of images in a perceptually consistent manner. Electronic industries are continuously moving forward with new products and technologies for digital cameras introduced in the market. Therefore it is necessary to have improved methods to verify that these technological advancements lead to higher quality images. This thesis reports the invention of a novel index for evaluation of image quality and its application for image compression and de-noising. The intended behavior of such metrics is to measure image quality as human observers would perceive it. The thesis first gives a background for image quality assessment, including the concept of image quality indexes. It was recognized that human vision was very sensitive to the distribution of brightness in an image; so here is an additional factor denoting, the shape of brightness histogram which is the difference between the grey-level histograms of the original and the distorted image into Structural Similarity Index Measure (SSIM). The proposed index, therefore, becomes a combination of four factors: luminance, contrast, structure and shape of brightness histogram. To ensure that the index can assist human observations in image quality evaluation, it needs to be assessed. A new method based on the ranking order was developed to assess the overall performance of image quality metrics. The proposed image quality index was assessed using existing traditional methods based on a set of public databases, which contains digital images of a range of different distortions, quality issues and quality ratings from human observers. The thesis then continues with some applications of the proposed image quality index in image compression and de-noising. It extends the commonly used algorithms for image compression and compares their performance. For image compression techniques, it has linked different wavelet techniques such as traditional mother wavelets and lifting based on Cohen-Daubechies-Feauveau wavelets with the lowpass filters of the length 9 and 7 (CDF 9/7),wavelet transform with Set Partition in Hierarchical Trees (SPIHT) algorithm and entropy coding. The main key point is to be demonstrated on the basis of proposed image quality index; the choice of mother wavelet becomes very important to achieve superior wavelet compression performances. An image de-noising, proposes a new de-noising algorithm based on Cohen-Daubechies-Feauveau (CDF 9/7) wavelet transforms. The results of proposed compression schemes show approximately 99% compression ratio and approximately 1 value of EIQ. It also proposes a new image de-noising approach using an optimum adaptive shrinkage threshold obtained with a proportional, integral and derivative (PID) tuning algorithm in the shearlet domain. The proposed denoising algorithm is more efficient for images contaminated with popular noises such as Gaussian noise, Poisson noise and impulse (salt & pepper) noise than all other methods. Experimental results show that images de-noised with the proposed approach have higher qualities than those produced with some of other de-noising methods like wavelet-based, bandlet-based, and curvelet-based using novel image quality indexes. The de-noised image is justified by EIQ and it’s approximately 1. The thesis also reports new algorithms, based on sparse, for image compression and de-noising. Compressed sensing (CS) is a new sampling theory that has recently been introduced for efficient acquisition of compressible signals. Compressed sensing (CS) states that if signals are sparse in some bases, then they will be recovered from a small number of random linear measurements via attractable convex optimization techniques. In this part of the thesis, these are two main contributions. In the first contribution, it is compressed images: high data throughput is becoming increasingly important in image, with high-resolution cameras (i.e., large numbers of samples per acquisition) and long observation times. The compressed sensing theory provides a framework to reconstruct images from fewer samples than traditional acquisition approaches. However, the very few measurements must be spread over a large field of view, which is difficult to achieve in conventional cameras. In this sense, it is proposes a novel method for compressive sensing of image reconstruction based on sparse representation of lifting scheme based on CDF 9/7 wavelet transform which is computational scheme to perform fast temporal acquisitions. In the second contribution, image de-noising: digital cameras images suffer from complex artifacts associated with low dark noise conditions. It has also proposed a novel image de-noising algorithm based on compressive sensing using multiple random under sampling in the sparse CDF9/7 wavelet domain and the Total Variation de-noising algorithm and adaptive Texture Variation with Adaptive Fidelity (ATVD) de-noising algorithm as a spatial sparsity prior for digital imaging. The experimental results of proposed image compression or de-noising method based on CS frameworks provide higher quality compressed and de-noised images compared the methods on open sources databases.
|Date of Award||2015|
|Supervisor||Xu Huang (Supervisor) & Kim Le (Supervisor)|