Research On Deep Learning Framework For Power Consumption Suppressing On Mobile Devices Display

Nowadays,smart mobile devices such as mobile phones and tablet computers have been widely used.Display is an indispensable interface for human-machine interaction in smart mobile devices and is divided into:Non-self-luminous displays and self-luminous displays.Organic Light-Emitting(OLED)as a modern emerging self-luminous display technology is different from the traditional non-luminescence display,and each pixel can provide a light source.The OLED brightness can be individually adjusted according to the content of the displayed image,and it is easy to effectively control the battery consumption.Power constrained contrast enhancement based on OLED self-luminous displays are a research hotspot.The traditional power-constrained image enhancement algorithm has two obvious deficiencies.First,the existing methods directly adjust the entire picture and affect the visual experience.Second,the degree of power saving is small.Therefore,how to achieve the power consumption without affecting the visual effects and smart mobile devices is a problem that we have studied and solved.Based on visual psychology,the decrease of the pixel brightness value in the image area in the display content has a greater impact on human visual perception,whereas the decrease in the pixel brightness value in the non-image area has a relatively small influence on visual perception.We divide the display content of smart mobile devices into image areas and non-image areas.Currently,the most advanced and widely used method of image semantic segmentation is Fully Convolutional Networks(FCN).However,when the FCN performs the semantic segmentation operation of the image region and the non-image region,there is a problem that the prediction image has a large area error and the edge division is not clear.We conducts the following two aspects of the study.(1)We have improved the FCN.A circular deep learning framework based on contextual regularization is proposed.This framework separates image and non-image areas.After experiment,the circular deep learning framework based on contextual regularization proposed in We has greatly improved the problem of large area errors and unclear edge segmentation of the FCN method.The four objective evaluation indicators,such as precision and recall rate,have been improved over the FCN.(2)Problems in Previous Power Saving Methods Based on OLED Displays,we propose a method of keeping the image area constant and reducing the pixel brightness value in non-image areas.According to the contextualized circular depth learning framework segmentation results,power saving operations are performed.This method can effectively maintain the image quality at the same time,making the display power consumption can be adjusted with the user's preferences.Experiments show that the proposed method is better subjectively and objectively than the other five most advanced power-saving algorithms based on OLED displays under the same power saving level.The method we propose can effectively maintain image quality and provide better visual experience.In addition,in this research process,we developed a software GTMaker which generates test image data sets.This software can not only serve the research of this topic,but also can serve similar research of more peers.