SAR Technology Image Recognition - Image File Details #1. Image Recognition - Upload Details. #2. Image Recognition - Flight Plans. #3. Image Recognition - Image Categories. #4. Image Recognition - User Registration. #5. Image Preparation - Resizing for Image Recognition. #6. Image Preparation - Segmentation for Image Recognition. #7 Image Recognition - Score Comments. #8. Image Recognition - Search Procedure (for registered users). -------------------------------------------------------------------- #1. Image Recognition - Upload Details. Still Images (jpg or png): Large, very high resolution (300 dpi or higher) images are preferred, especially for aerial images. Recommended min-max image size: - JPG & PNG: 200KB to 400KB e.g: 512 x 512 pixels (262,000 pixels) minimum .jpg files. - ** Copies of very large image files should be reduced to 400KB max before image recognition. ** - A maximum of 200 images are recommended for batch-recognition at one time. Video Images: Convert the videos to image files and then upload the still images for recognition: Conversion Example: Using 'Free Video to JPG Converter' https://www.dvdvideosoft.com/products/dvd/Free-Video-to-JPG-Converter.htm Original MP4 video file size: 3529 MB (13 minutes duration) Converts to 159 (0.5MB) .jpg files (Total 76 MB) at a 5 second image extraction interval. Converts to 266 (0.5MB) .jpg files (Total 152 MB) at a 3 second recommended image extraction interval. -------------------------------------------------------------------- #2. Image Recognition - Flight Plans. For aerial images i.e. from drones, helicopters and aircraft, the target may appear very small within the whole image. - Fly steadily at the very lowest safe altitude to make the target as large as possible within the images. - Take very high resolution (12MB, 4000 pixels wide by 3000 pixels tall) forward-looking still images (every 2-30 seconds), see Flight Plans below). - Use image stabilization and camera-shades to reduce blur and glare. - Turn off picture-taking between search-legs and when outside the search area. - Keep a careful record of the exact location where each image was taken (Set camera GPS Location On). - Record the date/time the aircraft starts and finishes each search area. Image Recognition Flight Plans, pre-configured for typical for typical aircraft, helicopter and drone-based searching can befound at: http://www.sartechnology.ca/imagerecognition-flightplans/ -------------------------------------------------------------------- #3. Image Recognition - Image Categories. 1. Crashed Aircaft or Shipwake: - Search by selecting the 'Crashed Aircraft' category. 2. Crashed Helicopters: - Search by selecting the 'Crashed Helicopter' category. 3. Persons or Vehicles: - Search by selecting the 'Persons' category. 4. Color Recognition: - Search by selecting the 'Color' Image Category. - The ‘Color’ category displays the image's dominant-colors scores. 5. Missing Vessels - General: - Search by selecting the 'General' Image Category. 6. General Image Recognition: - Keyword Search by selecting the 'General' Image Category. General Aircaft Keywords: Aeroplane | Air | Airbus | Aircraft | Airliner | Airplane | Airship | Aviate| Biplane | Engine | Flight | Flying | Fuselage | Hot Air Balloon | Jet | Parasail | Paraglider | Parachute | Propeller | Takeoff |Transportation System || Vehicle | Wing | General Helicopters Keywords: Copter | Helicopter |Transportation System | Vehicle | General Person Keywords: Adult | Backpack | Backpacker | Boy | Camper | Child | Climb | Climber | Climbing |Couple | Facial expression | Girl | Hike | Hiker | Hiking | Man | Mountain Climbing |People | Person | Persons | Woman | General Vehicle Keywords: Automotive |Car | Drag race | Drive | Jaunt |Transportation System | Travel | Trip (journey) | Truck | Vehicle | Wheel | General Vessels Keywords: Barge | Battleship | Boat | Boatman | Buoy | Canoe | Cargo Container | Cargo Ship | Catamaran | Clipper | Container Ship | Cruiser | Cruise Ship | Dinghy | Ferry | Fishing Boat | Fleet | Fisherman | Freighter | House Boat | Hull | Kayak | Life Jacket | Lifeboat | Motorboat | Navy | Oil Tanker | Passenger Ship | Pontoon | Sailboat | Schooner | Sail | Ship | Surfboarding | Surfboard | Transportation system | Tug | Tugboat | Vessel | Watercraft | Water Sports | Windsurf | Yacht | --------------------------------------------------------------------------- #4. Image Recognition - User Registration. To use SAR Technology's Image Recognition first register as a SAR Technology User: User Registration: From the SAR Technology Home page http://www.sartechnology.ca/ Goto Consultation... User Registration and register as a user: http://www.sartechnology.ca/user-registration/ Once your registration request has been received and approved you will be sent a temporary login password that will permit you to Login: http://www.sartechnology.ca/user-login/ You may then change your Password and User Profile: http://www.sartechnology.ca/user-profile/ ------------------------------------------------------------------------- # 5. Image Preparation - Resizing for Image Recognition. 1. Take pictures at the highest image resolution eg: 12 MB, 4000 pixels wide X 3000 pixels tall. Keep copies of these high-resolution full-size 12MB jpg images (in case they need to be segmented for zoom-in image recognition). 2. Batch-Resize for image recognition: Batch-Resize ** copies ** of the highest resolution (12MB) jpg images down to 1500 pixels wide X 1125 pixels tall (i.e. to approx 450-650 KB per jpg image) (e.g. Use Photoscape: http://www.photoscape.org/ps/main/index.php ) 3. Upload a known Target Image to the viewer (optional) Displays a known target image for comparison to the image recognition Search Results table (e.g for appearance or color). 4. Browse and upload 250-image batches of the 450-650 KB (approx) jpg images for image recognition. - Enter the image recognition %Threshold (optional) - leave blank to display all the uploaded images. - Enter the image-folder filepath (optional) - but useful for permanent documentation. Tip: A sample batch of image recognition SAR test images may be downloaded from: http://www.sartechnology.ca/sartechnology/MiscellaneousSARTestImages.zip 5. Click 'Predict image Files' and wait approx 1-3 minutes for the batch of images to be fully processed into the Search Results table. The uploaded images, their Image Recognition % Score and file Details will then be displayed in the Search Results table. The file Details include the image Filename, FileSize, Date/Time Created and the local Image-Folder Filepath (if entered). 6. Click the Search Results table's 'Score' header to sort by ranked % Score. (Click the browser's 'Wait' button(s) until the Search Results table is fully sorted by Score. 7. Print the ranked Score-sorted Image Recognition webpage (e.g. to a PDF file) to keep a permanent record of the image-recognition results. ------------------------------------------------------------------------- #6. Image Preparation - Segmentation for Image Recognition. Image segmenting (splitting) programs, such as Photoscape http://www.photoscape.org/ps/main/index.php may be used to segment these large-area images into smaller image segments e.g. into a 10x10 image grid of a 100 smaller sub-images. This will help the image recognition to correctly find the target within each sub-image. Image Segmentation Guidelines (for maximum aerial image recognition): http://www.sartechnology.ca/imagerecognition-segmentation/ Recommended Default Number of Segments: Persons: 72 Segments 12 Columns X 6 Rows (6X Zoom) Vehicles: 60 Segments 5 Columns X 12 Rows (5X Zoom) Aircraft, Helicopters & Vessels: 30 Segments 5 Columns X 6 Rows (6X Zoom) This 5-6X zoom-level number of segments generally provides the highest scores while minimizing the number of segments. Segments smaller than about 5-10KB may not create valid scores. Simultaneous Batch-Segmenting Procedure for large numbers of images: 1. Upload all of the original 12MB images into Photoscape - View screen. Select batches of 10 images and right-click Split into output subfolders, e.g. SegBatch1, SegBatch2 etc. ... Repeat this 10-image batch segmentation (splitting) until all of the original images have been segmented into batches. 2. Open each SegmentBatch subfolder in Windows Explorer... View... Medium-Icons... and expand the icon-display to full screen. Block-Select (Ctrl+MouseClick) an icon border and drag the selection rectangle across all the non-terrain images (e.g. blurred, sky etc), then delete these non-terrain images. Repeat deleting all non-terrain images from each segment subfolder. These terrain-only segment subfolders are now ready for uploading into the image-recognition webpage using #5. Image Preparation Step 4 - browse and upload - above. Because of their smaller size all of the (800-1200) subfolder segments may be uploaded at once. 3. Image-recognition search and then sort by Score the combined ten-batch folder results. Click all 'Wait' prompts, then print the webpage to the same batch-name PDF file. Use multiple Browser windows to increase throughput. (See tip below to Remove Firefox "Wait" prompts). 4. Use PDF Escape https://www.pdfescape.com/windows/ to rapidly view two PDF search reports at once. In Windows Explorer Right-click on the first PDF report and select Open with PDF Escape. Repeat for the second PDF report. Click the PDF-Escape View Tab to display and then select Side by Side View. Click the first document Select Path dropdown and select the first Opened Document. Repeat for the second document. Click OK to View these two PDF reports Side-by-Side. Click the left document, zoom to 1000% then drag until its first image-segment fills the left display. Repeat for the right document. Click the View Tab to display and then select Synchronous Scolling - to simultaneously scroll and review both PDF reports. 5. For fastest PDF viewing review only the top (high-scoring) 10 pages/ 5% of all the listed segment images. Typical Step 1-5 Image Processing Speeds (per person): 34.1 Original Images/Hour - when viewing only the high-scoring top 5% of all segments (fast). 16.5 Original Images/Hour - when viewing all of the displayed segments (thorough). Tip: Remove Firefox 'Wait" Prompts - when sorting large numbers (1,000) of Image-Recognition segment Scores i. Launch Firefox and in the browser’s address bar enter: "about:config" Press "Enter". ii. A Firefox warning page appears, stating that making changes to the advanced settings may be harmful. iii. Click “I accept the risk” to view the Settings... Preferences table. iv. In the settings page’s “Search” text box enter: "dom.max_script_run_time" When Firefox locates this preference, double-click the number "10" that appears in the table’s “Value” column and enter the new timeout (wait) value of "60" seconds. v. The new 60 second Firefox timeout (wait) delay is now in effect. Close the browser window. (Note: The 'Wait" prompt timeout value cannot be changed on the Chrome browser). --------------------------------------------------------------------------- #7 Image Recognition - Score Comments. (i) The higher the image relative-ranking position in the Score-sorted results list, the better the recognition confidence for the selected target category. (Persons, Vehicles, Crashed Aircraft, Shipwake or Crashed Helicopters). (ii) An Image Recognition relative-ranking in the top 10% of all sorted positions, or an absolute Score of 90% or more, is often a good confidence match for the selected target category. (iii) Higher quaility, sharper images create much higher %Score values than similar blurred, grainy, out-of-focus or low-contrast images. (iv) Score Ranking sorts both good-quality and poor-quality image batches to rank the target object towards the top of the ranked list. (v) Higher absolute %Score values can be considered as having higher confidence, while lower absolute %Score value images can be considered as having a lower confidence, in identifying the target object. --------------------------------------------------------------------------- #8. Image Recognition - Search Procedure (for registered users). Goto SAR Technology... Mission Services... Image Recognition-Login webpage: http://www.sartechnology.ca/imagerecognitionuserlogin/ On the Image Recognition webpage: Select the required image category. - Crashed Aircraft - Crashed Helicopters - Missing Persons - Color - Vessels - General -------------------------------------------------------------------------