Video Compression using Neural Weight Step and Huffman Coding Techniques

Background:
 This paper proposes a Hierarchical Video Compression Scheme (HVCS) method with three hierarchical layers of quality Recurrent Quality Enhancement (RQEN‎‎) network. Image compression techniques are used to compress frames in the first layer, where have highest quality. Using high-quality frame as reference, Bi-Directional Deep (BDC) network is proposed for second layer considerable In third low using adopted Single Motion (SMC) network, which single motion map estimation within multiple frames. As result, SMC provide information fewer bits. decoding stage, weighted developed take both bit stream and compressed inputs. RQEN cell‎‎, update signal memory features positively influence multi-frame enhancement. this paper, HVCS adopts benefit efficiency coding, whereas improves enhances low-quality at encoding stages, respectively. Experimental results validate that approach overcomes state-of-the-art methods.
 Materials Methods:
 Tables 1& 2 illustrate representing yielded values rate-distortion on video datasets. aforementioned, PSNR MS-SSIM evaluation, bit-rates calculated bits per pixel (bpp). Table 1 illustrates performance, they show better performance model than other methods such Chao et al [7] or optimized [1]. addition, outperform applying H.265 standard JCT-VC dataset. On side, scheme bit-rate UVG. 2, evaluation provided all learned approaches, it reached H.264 H.265. Due UVG, Lee ‎et al. [11] has comparable Guo [10] lower Applying JCT-VC, DVC only opposite, preformance HVCS-rate-distortion obvious H.265.
 Furthermore, BjꝊntegaard Delta Bit-Rate (BDBR) [47] also computed depending A BDBR measure computes average difference considering anchor, indicated [48]. 3, illustrated MS-SSIM, which, reduction anchor by showed negative numbers. Such bold numbers represent best methods. 3 fair comparison (MS-SSIM & PSNR) 
 Results:
 Conclusion:
 work utilizing recurrent Specifically, dividing into levels 1, decreasing For image proposed, while proposing BDC RQEN‎‎ inputs validated scheme.
 Similarly techniques, structure manually set scheme. promising recommendation future can be accomplished developing DNN networks automatically prediction hierarchy.