Dishan Technology Establishes Image Sensor Noise Reduction Technology: Delivering Breakthrough Imaging Experiences Through an Unquenchable Thirst for Exploration
Today, with the rapid development of digital imaging technology, imaging quality has become a crucial benchmark for measuring scientific and technological strength. Currently, the image sensors of certain high-end smartphones on the market have achieved a resolution of 100-megapixel level. Autonomous vehicles rely on high-precision imaging to enable safer driving; the medical field uses high-resolution imaging for accurate diagnosis; and the aerospace sector depends on image sensors for Earth observation and space exploration.
These examples demonstrate that from smartphones to autonomous driving, and from medical imaging to aerospace, image sensors—dubbed the "eyes of machines"—directly determine the accuracy and reliability of information acquisition. In this critical field, Dishan Technology has established industry-leading technological barriers through its in-depth research into noise reduction technology. Driven by an unquenchable spirit of exploration, it has redefined the boundaries of high-fidelity imaging, delivering unprecedented breakthrough imaging experiences to users worldwide.
Behind this achievement lie countless experiments and breakthroughs, the constant challenging of technological limits, and above all, an unwavering pursuit of the goal—to see the true nature of the world.
The Difficulty of Noise Reduction: The 'Invisible Killer' of Imaging Quality
In complex shooting environments such as low light, high sensitivity, or long-term exposure, image sensors are prone to generating noise - chaotic particles, color deviations, and blurry details that not only weaken the purity of the image, but also seriously affect the accuracy of subsequent image recognition and analysis. The presence of noise, like "noise" mixed in during information transmission, makes it difficult for machines to accurately capture real visual information. Traditional noise reduction methods often rely on post-processing algorithms to smooth and filter images through software. However, this method often sacrifices details, resulting in a "grainy feeling" in the image, loss of true texture, blurred edges, and even artifacts. For example, in surveillance footage captured at night, traditional noise reduction may result in blurred facial features and difficulty in recognizing license plate numbers; In medical imaging, the loss of tissue texture may affect doctors' diagnostic judgments. The contradiction of how to suppress noise while preserving original details is like the two ends of a balance, which has always been the "hard bone" in the field of imaging technology.
Technological breakthrough: full chain innovation from pixel structure to AI collaboration
The noise reduction technology of Dishan Technology is not a single technological breakthrough, but a systematic and full chain innovation, covering multiple dimensions such as sensor hardware architecture, signal processing, algorithm optimization, etc., forming a complete noise reduction solution.
New Pixel Structure Design: Improving Light Signal Quality from the Source
Traditional pixel structures have low photon capture efficiency in low light environments, resulting in weak signals and significant noise. Di Shan Technology has designed a unique "honeycomb shaped" micro lens array by optimizing the structure of photodiodes and the layout of micro lenses, increasing the photon receiving area of each pixel by 20% while reducing the loss of optical signals during transmission. In addition, the use of new semiconductor materials reduces dark current noise and physically reduces the generation of noise. This improvement not only improves the signal-to-noise ratio, but also enhances the imaging capability of the sensor in low light environments, making "high signal-to-noise ratio acquisition" possible. Specifically, the team developed a pixel layer based on gallium arsenide (GaAs) and graphene composite materials, utilizing the high conductivity of graphene and the photoelectric conversion efficiency of GaAs to achieve a significant improvement in quantum efficiency. In laboratory testing, the new pixel can still output usable images under 0.1 lux illumination, while traditional sensors have completely failed in this environment.
Dual gain parallel readout architecture: dynamically balancing signal and noise
Integrate a dual channel signal reading mechanism inside the sensor to dynamically switch between high gain and low gain channels. High gain channels are suitable for low light environments and can amplify weak signals; Low gain channels are used in high light scenes to avoid signal overflow. Real time judgment of scene brightness through intelligent algorithms, dynamic adjustment of gain settings, effectively expanding the dynamic range to over 120dB, and suppressing highlight overflow and dark noise. This innovation breaks the dilemma of traditional sensors being "high gain, high noise, low gain, and low dynamics", achieving dynamic balance between signal and noise. For example, when shooting portraits in the noon sun with backlighting, the sensor can simultaneously preserve the details of the sky and the texture of the person's face, avoiding overexposure or underexposure.
On chip AI noise reduction engine: real-time intelligent noise reduction
Embedding lightweight deep learning models into sensor ISP (Image Signal Processor) to complete noise feature recognition and real-time filtering at the moment of data acquisition. AI models trained on massive amounts of data can accurately distinguish between noise and real details, such as distinguishing fabric textures from noisy particles, while preserving subtle structures such as hair strands and leaf veins. Compared with traditional algorithms, AI denoising is more intelligent, accurate, and has extremely low latency, achieving "noise reduction while collecting" and avoiding image quality loss caused by post-processing. This technology is particularly suitable for scenarios that require high real-time performance, such as autonomous driving, motion capture, etc. The team has developed a dedicated neural network architecture that compresses the model to only 1% of the chip's computing power while maintaining 98% noise reduction accuracy. This breakthrough benefited from the "edge computing Joint Laboratory", which cooperated with colleges and universities. Through the collaborative optimization of algorithms and hardware, the balance between performance and power consumption was achieved.
Space time domain joint denoising algorithm: multi-dimensional denoising, precise restoration
By combining multi frame temporal information and spatial texture analysis, noise is identified by comparing the changes between consecutive frames, while using the spatial correlation of the image to distinguish between real texture and random noise. For example, when shooting motion scenes, multi frame fusion technology is used to eliminate motion blur while suppressing fixed noise; In static scenes, utilizing the correlation between adjacent pixels to restore details. This algorithm avoids the excessive smoothing caused by traditional noise reduction algorithms' one size fits all approach, achieving precise separation of noise and details. During the development process, the team collected over 10TB of noise sample library, covering data under different lighting, temperature, and motion states, and continuously optimized the algorithm's generalization ability through machine learning. In the end, the algorithm was able to preserve over 90% of image details at ISO 12800 high sensitivity, far exceeding industry standards.
The insatiable desire for exploration: the soul driving force behind innovation
Technological breakthroughs stem from spiritual drive. The Dishan Technology team always adheres to the exploration belief of "never satisfied" - not satisfied with existing performance indicators, not satisfied with industry standard solutions, and not satisfied with compromises of "enough is enough". They believe that true technological leadership comes from the ultimate polishing of every tiny detail and the continuous challenge of the impossible. This spirit runs through every aspect of research and development:
The insatiable desire for exploration: the soul driving force behind innovation
Technological breakthroughs stem from spiritual drive. The Dishan Technology team always adheres to the exploration belief of "never satisfied" - not satisfied with existing performance indicators, not satisfied with industry standard solutions, and not satisfied with compromises of "enough is enough". They believe that true technological leadership comes from the ultimate polishing of every tiny detail and the continuous challenge of the impossible. This spirit runs through every aspect of research and development:
Breakthrough Imaging Experience: Empowering Thousands of Industries with a Visual Revolution
The noise reduction technology of Dishan Technology not only serves to improve image quality, but also brings about changes in multiple fields, driving the industry towards higher precision and intelligence
Future Eye: Illuminate a New Era of Imaging with the Light of Exploration
Establishing a leading position in noise reduction technology is not the end, but a new starting point. Di Shan Technology is extending this technology to more cutting-edge fields and continuously exploring the boundaries of imaging technology
