Wireless video transmission and video compression technology are the foundations and cores of many Internet of Things(IoT)applications and embedded systems in these days. However,multi-channel transmission always causes video frame loss and delay jitter because of the continuous change of wireless network state. Although the adaption algorithm can solve the video transmission problem under PC or server platform to a certain extent,the real-time performance and Quality of Service(QoS)requirement cannot be satisfied under the embedded platform and wireless network. Therefore, based on the DM368 chip,a complete platform was designed from video capture,compression,WiFi transmission,control unit reception to host computer display. At the same time,with the full consideration of the characteristics of embedded platform,a rate adaption algorithm that combines signal quality,network bandwidth,buffer status and congestion control was proposed. In this algorithm,the Gaussian function was used to calculate network bandwidth,the segmented inverse proportional function was used to adjust buffer status,the weighted moving method was adopted to smooth rate,and the extreme value suppression method was used for rate balancing. The smooth rate adjustment was realized by this algorithm, and the algorithm was applied to the proposed platform to realize the management of the control unit on multiple WiFi cameras,multi-channel transmission and load balancing. The QoS was used as the evaluation index for experimental verification. The results show that the algorithm performs well on the embedded platform with great improvements of smoothness and buffer stability,and has significantly fairness and bandwidth utilization improvements under multi-channel condition. In a variety of situations,such as single camera signal quality dynamic change or multi-camera bandwidth competition,compared with the McGinely Dynamic Indicator(MDI)algorithm,the proposed algorithm has the smoothness improved by 16% to 59%;compared with the Buffer-Based Algorithm(BBA),the proposed algorithm has the cache jitter reduced by 15% to 72%,and the delay jitter reduced by 12% to 76%.

Existing weakly illuminated image enhancement algorithms are strongly dependent on Retinex model and require manual adjustment of parameters. To solve those problems, an algorithm based on Convolutional Neural Network (CNN) was proposed to enhance weakly illuminated image. Firstly, four image enhancement techniques were used to process weakly illuminated image to obtain four derivative images, including contrast limited adaptive histogram equalization derivative image, Gamma correction derivative image, logarithmic correction derivative image and bright channel enhancement derivative image. Then, the weakly illuminated image and its four derivative images were input into CNN. Finally, the enhanced image was output after activation by CNN. The proposed algorithm can directly map the weakly illuminated image to the normal illuminated image in end-to-end way without estimating the illumination map or reflection map according to Retinex model nor adjusting any parameters. The proposed algorithm was compared with Naturalness Preserved Enhancement Algorithm for non-uniform illumination images (NPEA), Low-light image enhancement via Illumination Map Estimation (LIME), LightenNet (LNET), etc. In the experiment on synthetic weakly illuminated images, the average Mean Square Error (MSE), Peak Signal-to-Noise Ratio (PSNR) and Structural SIMilarity index (SSIM) metrics of the proposed algorithm are superior to comparison algorithms. In the real weakly illuminated images experiment, the average Natural Image Quality Evaluator (NIQE) and entropy metric of the proposed algorithm are the best of all comparison algorithms, and the average contrast gain metric ranks the second among all algorithms. Experimental results show that compared with comparison algorithms, the proposed algorithm has better robustness, and the details of the images enhanced by the proposed algorithm are richer, the contrast is higher, and the visual effect and image quality are better.

For image instance retrieval in the field of computational vision, a semantic region weighted aggregation method based on significance guidance of deep convolution features was proposed. Firstly, a tensor after full convolutional layer of deep convolutional network was extracted as deep feature. A feature saliency map was obtained by using Inverse Document Frequency (IDF) method to weight deep feature, and then it was used as a constraint to guide deep feature channel importance ordering to extract different special semantic region deep feature, which excluded interference from background and noise information. Finally, global average pooling was used to perform feature aggregation, and global feature representation of image was obtained by using Principal Component Analysis (PCA) to reduce the dimension and whitening for distance metric retrieval. The experimental results show that the proposed image retrieval algorithm based on significant semantic region weighting is more accurate and robust than the current mainstream algorithms on four standard databases, because the image feature vector extracted by the proposed algorithm is richer and more discerning.

To solve the problem that the text regions can not be extracted well in low contrast images by traditional Maximally Stable Extremal Regions (MSER) method, a novel scene text detection method based on edge enhancement was proposed. Firstly, the MSER method was effectively improved by Histogram of Oriented Gradients (HOG), the robustness of MSER method was enhanced to low contrast images and MSER was applied in color space. Secondly, the Bayesian model was used for the classification of characters, three features with translation and rotation invariance including stroke width, edge gradient direction and inflexion point were used to delete non-character regions. Finally, the characters were grouped into text lines by geometric characteristics of characters. The proposed algorithm's performance on standard benchmarks, such as International Conference on Document Analysis and Recognition (ICDAR) 2003 and ICDAR 2013, was evaluated. The experimental results demonstrate that MSER based on edge enhancement in color space can correctly extract text regions from complex and low contrast images. The Bayesian model based classification method can detect characters from small sample set with high recall. Compared with traditional MSER based method of text detection, the proposed algorithm can improve the detection rate and real-time performance of the system.

Focusing on the problem of object structure discontinuity and poor texture detail occurred in image inpainting, an inpainting algorithm based on self-adaptive group was proposed. Different from the traditional method which uses a single image block or a fixed number of image blocks as the repair unit, the proposed algorithm adaptively selects different number of similar image blocks according to the different characteristics of the texture area to construct self-adaptive group. A self-adaptive dictionary as well as a sparse representation model was established in the domain of self-adaptive group. Finally, the target cost function was solved by Split Bregman Iteration. The experimental results show that compared with the patch-based inpainting algorithm and Group-based Sparse Representation (GSR) algorithm, the Peak Signal-to-Noise Ratio (PSNR) and the Structural SIMilarity (SSIM) index are improved by 0. 94-4.34 dB and 0. 0069-0.0345 respectively; meanwhile, the proposed approach can obtain image inpainting speed-up of 2.51 and 3.32 respectively.

Defect detection is an important part of fabric quality control. To make the detection algorithm possess good commonality and high detection accuracy, a dual-scale fabric defect detection algorithm based on sparse coding was proposed. The algorithm combined the advantage of high stability under large-scale and the advantage of high detection sensitivity under small-scale. At first, the dictionaries under large and small scales were obtained through a small-scale over-complete dictionary training method. Then, the projection of detection image block on the over-complete dictionary was used to extract detection characteristics. Finally, the detection results under dual-scale were fused by the means of distance fusion. The algorithm overcame the disadvantage of large computation because of the introduction of dual-scale while using small-scale over-complete dictionary and downsampling the detection image under large-scale. TILDA Textile Texture Data base was used in the experiment. The experimental results show that the algorithm can effectively detect defects on plain, gingham and striped fabric, the comprehensive detection rate achieves 95.9%. And its moderate amount of calculation can satisfy the requirement of industrial real-time detection, so it does have much value of practical application.