Wind Yuan | a7fe91e | 2017-07-10 11:15:08 -0400 | [diff] [blame] | 1 | 2017/07/10: release libxcam version 1.0.0 |
| 2 | * 360 video stitching performance and quality improvement. |
| 3 | - enable geometry map to improve performance. |
| 4 | - quality tuned on different resolutions (1080P and 4K). |
| 5 | - support CPU and OpenCL path in feature match. |
| 6 | - enable lens shading correction based on fisheye image. |
| 7 | |
| 8 | * gyroscope-based video stabilization enabling. |
| 9 | - enable gyroscope 3-DoF (orientation) assist video stabilization. |
| 10 | - orientation (gyro) data should be measured by quaternion as the pose |
| 11 | of target frame reference to base frame |
| 12 | |
| 13 | * CL framework refine. |
| 14 | - enable CL argument template instead of member variables in kernel. |
| 15 | - CL kernel support async mode. |
| 16 | - image handler take over input/output buffer management from image kernel. |
| 17 | |
| 18 | * prepare libxcam debian package and fix most warnings. |
| 19 | |
Wind Yuan | a682b9e | 2017-04-01 06:40:59 -0400 | [diff] [blame] | 20 | 2017/04/01: release libxcam version 0.9.0 |
| 21 | * 360 video stitching improvement. |
| 22 | - remove sharp boundary in sphere mode by local/global area resize. |
| 23 | - tune algorithm parameters to preserve blinking. |
| 24 | - integrate 360 stitching pipeline into libxcam module. |
| 25 | - gst-xcam-filter supports 360 stitching case. |
| 26 | |
| 27 | * export C APIs for each OpenCL feature. |
| 28 | - support stich/3d-nr/wavelet-nr/defog features by xcam handle. |
| 29 | - be designed for FFmpeg xcam video filter usage. |
| 30 | |
zongwave | aeedc8f | 2017-01-03 15:27:11 +0800 | [diff] [blame] | 31 | 2016/12/30: release libxcam version 0.8.0 |
| 32 | * Digital Video Stabilization (DVS) |
| 33 | - enable DVS smart analysis plugin. |
| 34 | - algorithms used in DVS (feature detect, optical flow, motion model estimation) are based on OpenCV. |
| 35 | - implement image warp perspective algorithm in CL kernel. |
| 36 | |
| 37 | * Image stitching to create panorama photography |
| 38 | - implement fisheye calculation which map location to equirectangular. |
| 39 | - enable feature detection and match based on OpenCV optical flow algorithm. |
| 40 | - enable seam mask to improve the quality of blender. |
| 41 | |
| 42 | * gstreamer plugin xcamfilter can co-work with xcamsrc |
| 43 | |
Wind Yuan | 2f6d8b5 | 2016-09-29 03:37:53 -0400 | [diff] [blame] | 44 | 2016/09/30: release libxcam version 0.7.0 |
| 45 | * 3D-NR quality and performance improvement. |
| 46 | - use image exposure parameters to adjust global denoise strength. |
| 47 | - calculate image local variation to adjust local denoise strength. |
| 48 | - optimize OCL kernel by shared local memory and sub-group to |
| 49 | improve performance. |
| 50 | |
| 51 | * defog/dehaze quality improvement based on DCP(dark channel prior). |
| 52 | - enable dark channel prior to improve hazy image quality. |
| 53 | - fix image corruption in recover step. |
| 54 | - add lateral-range filter to smooth dark channel and remove |
| 55 | halo effect. |
| 56 | |
| 57 | * enable multi-band blender for image stitching. |
| 58 | - based on gaussian and laplacian pyramid and blend images in each |
| 59 | band, then reconstruct stitching image from pyramid level. |
| 60 | - blender can smooth seams and prevent ghosting. |
| 61 | |
| 62 | * enable geometry correction. |
| 63 | - map 2D image from src to dst image based on coordinate table which |
| 64 | is center aligned. |
| 65 | - prevent sawtooth edge by OCL sampler. |
| 66 | * gst-xcamfilter feature supports. |
| 67 | - xcamfilter can co-work with different camera source. |
| 68 | |
Wind Yuan | 7d4b6a8 | 2016-07-29 20:19:20 +0800 | [diff] [blame] | 69 | 2016/07/29: release libxcam version 0.6.1 |
| 70 | * enable 3D noise reduction, performance stay tuned. |
| 71 | - suppress image noise in both temporal and spatial domain. |
| 72 | - the algorithm applies denoise filter based on pixel-block differences |
| 73 | in neighbor blocks and previous frames. |
| 74 | - the processed image is much more clean and preserves motion objects |
| 75 | edge in low-light conditions. |
| 76 | |
| 77 | * support gst xcamfilter plugin which is independent from camera source. |
| 78 | - xcamfilter plugin can work with gstreamer source plugins, |
| 79 | e.g. videotestsrc/filesrc/v4l2src. |
| 80 | - xcamfilter features support defog/wireframe/waveletNR/3d-denoise. |
| 81 | - sink pad supports format NV12, src pad supports DMA buffer-sharing |
| 82 | for performace if links to HW plugin(encoder). |
| 83 | - wireframe/scaler/waveletNR moved to post image processor which can |
| 84 | co-work with isp processor(as pre-image processor) |
| 85 | |
Wind Yuan | 16cac07 | 2016-07-01 03:33:29 +0800 | [diff] [blame] | 86 | 2016/06/30: release libxcam version 0.6.0 |
| 87 | * enable block-split version of histogram based WDR tone-mapping algorithm. |
| 88 | - divide image into several blocks and apply bidirectional log based |
| 89 | function on luminance to improve the contrast inside each block. |
| 90 | - apply weight function based on distance between target pixel and center |
| 91 | pixel of each neighbor block to smooth and eliminate block boundaries. |
| 92 | - apply weight function based on pixel intensity deviation between neighbor |
| 93 | blocks keep differentiation and weaken boundaries. |
| 94 | - image quality is improved and more details in both over-exposured and |
| 95 | under-exposured area can be shown clearly on screen. |
| 96 | |
| 97 | * Bayesian wavelet shrinkage for adaptive noise filtering. |
| 98 | - enable adaptive noise filtering based on Bayesian shrinkage estimation |
| 99 | - estimate the image’s background noise level by analyzing. |
| 100 | wavelet coefficient. |
| 101 | - optimize the threshold T which minimizes the Bayesian risk, |
| 102 | i.e, the expected value of the mean square error. |
| 103 | |
| 104 | * Haar wavelet tuning improvement. |
| 105 | - fine tuning the threshold statistic data for ultra-low light condition. |
| 106 | to prevent edge from over smoothing. |
| 107 | |
| 108 | * face detection supports. |
| 109 | - enable face detection framework and support FD plugins. |
| 110 | - enable wire-frame to track faces. |
| 111 | |
| 112 | * fog removal improvement based on Retinex algorithm. |
| 113 | - support multi-scale algorithm to reduce halo effects. |
| 114 | - tune parameters to keep brightness and clear edge. |
| 115 | - improve saturation based on chroma enhancement. |
| 116 | |
| 117 | * image processing service framework. |
| 118 | - add pipe-manager to support smart-analysis and post image processor. |
| 119 | - separate major code into different modules, e.g. xcore/isp/ocl/3a. |
| 120 | - support NV12 stream as fake input instead of camera sensor. |
| 121 | |
Wind Yuan | e48099f | 2016-03-23 22:44:10 +0800 | [diff] [blame] | 122 | 2016/03/28: release libxcam version 0.5.0 |
| 123 | * enable new WDR tone mapping based on histogram adjustment. |
| 124 | - apply bidirectional log based function on luminance to improve |
| 125 | image contrast. |
| 126 | - transmit wide dynamic range image data to 8-bit color data by |
| 127 | novel histogram adjustment algorithm. |
| 128 | - details in both over-exposured and under-exposured area clearly |
| 129 | show up on screen. |
| 130 | * enable wavelet-based algorithm for noise reduction. |
| 131 | - decompose image into multiple-scales by wavelet transform with |
| 132 | low pass and high pass filters. |
| 133 | - perform soft threshold on smaller coefficients(high-pass) |
| 134 | to reduce noise. |
| 135 | - threshold should be suitably decided by tuned parameters. |
| 136 | - reconstruct image from scaled low-pass and high-pass images. |
| 137 | - enable Haar-wavelet and Hat-wavelet NR filters. |
| 138 | * enable fog removal feature based on retinex algorithm. |
| 139 | - base on retinex single-scale algorithm. |
| 140 | - scale down image for gaussian blur to improve performance. |
| 141 | - scale up blured image and amplify log differences with original |
| 142 | image. |
| 143 | * improve sharpness in normal light with extreme profile. |
| 144 | * support cl post processing with isp mode. |
| 145 | * support swap-buffer on nv12 format if only single plane need to |
| 146 | process. |
| 147 | |
Wind Yuan | c5f78c0 | 2015-12-30 23:49:42 +0800 | [diff] [blame] | 148 | 2015/12/30: release libxcam version 0.4.0 |
| 149 | * improve performance on OCL pipeline. |
| 150 | - design new formats and pipeline to save memory IO. |
| 151 | - improve bayer noise reduction with bilateral filter. |
| 152 | - support edge enhancement into bayer pipe. |
| 153 | - make gpu/cpu working together for 3a-statistics calculation. |
| 154 | - move tonemapping kernel from RBG to bayer format. |
| 155 | * improve WDR-tonemapping algorithm for quality. |
| 156 | - utilize an new adaptive local tone mapping algorithm. |
| 157 | - provide the mapping between 16-bit wide dynamic range images to |
| 158 | 8-bit images to have “nice looking”. |
| 159 | - preserve more information on luminance values of the scene |
| 160 | especially for high contrast images. |
| 161 | * tonemapping parameters calculated to increase the contrast between |
| 162 | high light and low light. |
| 163 | - tone mapping regularization parameters are calculated according |
| 164 | to the distribution of brightness histogram. |
| 165 | - the local adaptive parameter are calculated by weighting the |
| 166 | neighbor pixels with a Gaussian blur filter. |
| 167 | * add bayer raw video input to simulate camera sensor. |
| 168 | |
Wind Yuan | 37f60d5 | 2015-10-24 03:45:20 +0800 | [diff] [blame] | 169 | 2015/10/19: release libxcam version 0.3.0 |
| 170 | * improved OCL pipeline on basic and advanced/extreme pipeline |
| 171 | - add new bayer noise-reduction into demosaic |
| 172 | - merged yuv-tnr into yuv-pipe kernel |
| 173 | - merged TNR-yuv and TNR-rgb together |
| 174 | * support WDR(wide dynamic range) feature |
| 175 | * add 3a analysis tuner framework for more features tunning, e.g TNR |
| 176 | * add hybrid 3a analysis framework |
| 177 | - support partial customized 3a algorith, e.g AWB/AE |
| 178 | * add smart analysis framework |
| 179 | - generate small scaled picture |
| 180 | - support customized to analyze small picture and feedback results |
| 181 | - support loading user-defined method on smart analysis |
| 182 | |
Wind Yuan | 0d7d28c | 2015-07-30 17:53:06 +0800 | [diff] [blame] | 183 | 2015/07/31: release libxcam version 0.2.1 |
| 184 | * improved OCL pipeline on basic 3a image processing |
| 185 | - merged bayer kernels with blc, wb, gamma, demosaic and 3a stats. |
| 186 | - merged yuv kernerls with rgb2yuv color conversion and macc. |
| 187 | - support async framework as option on cl features |
| 188 | - support OCL buffer allocation which can choose tiling mode. |
| 189 | * support different AIQ versions. |
| 190 | * fix capture of dead-loop and sensor format setting. |
| 191 | |
Wind Yuan | 47b5e4d | 2015-07-02 13:13:15 +0800 | [diff] [blame] | 192 | 2015/07/02: release libxcam version 0.2.0 |
| 193 | * add gstreamer plugin 'xcamsrc' for Linux media framework support |
| 194 | * support loading user-defined 3a algorithm lib dynamically |
| 195 | * support more manual 3a features on ISP, e.g multiple-ae-window, ae/awb speed... |
| 196 | * add OpenCL pipeline for 3a image processing |
| 197 | - bayer format features, black level correction, defect pixel |
| 198 | correction, whitebalance, 3a statistics calculation, |
| 199 | demosaic, gamma correction, HDR(high dynamic range) |
| 200 | - RGBA format features, bilateral NR(noise reduction), simple NR, |
| 201 | temporal NR, macc, color correction |
| 202 | - YUV format features, temporal NR, edge enhancement, color conversion |
| 203 | - support any user-defined 3a algorithms(e.g AIQ) |
| 204 | * rich test cases, e.g. test-device-manager, test-cl-image |
| 205 | |
Wind Yuan | 75564b1 | 2015-01-15 06:51:15 -0500 | [diff] [blame] | 206 | 2015/01/15: Initial xcam version 0.1 |