Publications
Image compression based on frequency domain reduction size
Dec 15, 2023Publisher 1ST INTERNATIONAL CONFERENCE ON SUSTAINABLE DEVELOPMENT TECHNIQUES (ICSDT2022
DOI https://doi.org/10.1063/5.0171385
Issue 1
Volume 2862
This work proposes a new image compression method based on a DCT (Discrete Cosine Transformed) combined with the Matrix Size Reduction algorithm. The compression algorithm starts by dividing the image into 8x8 blocks, then DCT is applied to each block independently, followed by uniform quantization. After that, a zigzag scan is applied to each block to be a one-dimensional array. Additionally, the array size is reduced by eliminating insignificant coefficients using the Matrix size-reduced algorithm. Afterward, the residual coefficients are compressed by Arithmetic Coding. The Matrix Reduction size algorithm is accomplished based on two different random keys. And then, two adjacent frequency domain coefficients are reduced to a single value. The decompression uses a searching method called Sequential Search Algorithm to decode the previously compressed data to retrieve residual coefficients. These residuals are padded with zeros to rebuild the original 8x8 blocks. Finally, inverse DCT is applied to reconstruct approximately the original image. The experimental results showed that our proposed image compression and decompression have achieved up to 98% compression ratio, keeping most of the visual image quality
THE AUTOMATIC LICENSE PLATE RECOGNITION USING FEATURES EXTRACTION AND NEURAL NETWORKS
Dec 8, 2020Journal Journal of University of Duhok
Publisher 3rd international conference on recent innovations in engineering (ICRIE) Duhok,
Volume 23
The automatic license plate recognition (ALPR) system opens the trendy door to the researchers to think, discover techniques and reach to a result for its necessity. The important of the ALPR system is appeared in the transportation for many reasons such as parking, traffic violations and security. The aim of this paper is to suggest a scheme that will extract car number, country and province from the car images. The proposed scheme is based on digital image processing techniques and neural networks. The proposed algorithm is composite of preprocessing and recognition stages. The preprocessing stage includes: locate the car plate region, binarization, enhancement of the image quality, segment the image into the sub-images. The recognition stage will classify and recognize the segmented sub-images as numbers and characters. In this research, the localization is done through normal cross correlation method. The segmentation includes: segment the car plate into three regions, divide the number and separated character into individual and split the connected characters into separated characters are done through suggested algorithms. The recognition is accomplished using the back propagation neural network (BPNN). The recognizer operates on two sets of data. First set of data includes the whole pixels of the sub-images. The second set of data is based on 16 features extracted from the sub-images. A comparison between these two methods is made. The system is experienced on 99 images of Duhok and Erbil provinces, the environment work is done with MATLAB program. The percentage accuracy is: 100%,100% and 100% for the localization, distinguish and segmentation respectively. The recognition rate result for the first method is 94.5% and the second method is 91%.